How do AI agents pay for APIs on TensorFeed?

Agents pay in USDC on Base mainnet. There are no accounts, no API keys, and no traditional payment processors. An agent buys credits once via /api/payment/buy-credits, sends USDC to the published wallet, calls /api/payment/confirm with the tx hash, and receives a bearer token that debits credits per premium API call.

What is x402 and is TensorFeed compatible?

x402 is the emerging HTTP standard for machine-payable APIs: the server returns HTTP 402 Payment Required with payment instructions, and the client retries with proof of payment. TensorFeed supports both the credits-first flow (recommended for repeat use) and an x402-style fallback where a single call without auth returns a 402 with a wallet address, and a retry with the X-Payment-Tx header serves the data.

Why USDC on Base instead of Stripe or PayPal?

AI agents do not have credit cards or human-mediated checkout. USDC on Base settles in seconds with no chargebacks, no KYC at the API layer, and no platform risk from a payment processor. Base is an Ethereum L2 with sub-cent gas fees, making microtransactions ($0.02 per credit) viable. We accepted a smaller short-term TAM (only crypto-native agents) for zero processor fees and cleaner agent-native positioning.

How do I verify the TensorFeed payment wallet address before sending funds?

Cross-check the wallet address across four published locations before sending USDC: /llms.txt on tensorfeed.ai, the JSON response from /api/payment/info, the README in github.com/RipperMercs/tensorfeed, and the @tensorfeed bio on X. If any of the four disagree, do not send. The address is also published with TLS so a network MITM cannot swap it.

What happens if my payment fails or the API rejects the transaction?

On-chain verification reads the USDC Transfer event from eth_getTransactionReceipt on the Base RPC. If the tx is invalid, the wrong recipient, or already used, /api/payment/confirm returns an error and no credits are minted. Replay protection means every successful tx hash is permanently recorded so the same payment cannot be claimed twice. Note that confirmed credit purchases are non-refundable per Section 17.5 of the Terms; credits do not expire, so the recommended pattern is to buy small at first and top up as call volume is calibrated.

Are TensorFeed premium APIs covered by an SLA?

No SLA in the MVP phase. Premium credits do not expire and there is no scheduled downtime, but if a premium endpoint is down we will not credit your account automatically. Credit purchases are non-refundable; the practical mitigation is the small-purchase pattern (start at $1 USDC for 50 credits, top up after calibration). As the product matures we may introduce paid SLA tiers.

Can I use TensorFeed premium data to train my AI model?

No. Premium data is licensed for inference use only. Use of TensorFeed premium endpoint responses for training, fine-tuning, evaluation, or distillation of machine learning models is prohibited. This is in the Terms of Service section on Premium API and Agent Payments.

Does TensorFeed work with the Model Context Protocol (MCP)?

Yes. The @tensorfeed/mcp-server package exposes every premium endpoint as an MCP tool, so Claude Desktop, Claude Code, and other MCP clients can call routing, history series, watches, and the enriched agents directory directly. Pass your tf_live_ bearer token via the TENSORFEED_TOKEN env var in your MCP client config.

Agent Payments API

Premium API access for AI agents. Paid in USDC on Base, settled in seconds. No accounts, no API keys, no traditional payment processors.

Send USDC, get a bearer token, call premium endpoints. Each call decrements credits from your token. When credits run out, top up. The token is the only credential the API ever sees from you.

Every paid call is Agent Fair-Trade certified.

AFTA

Code-enforced no-charge on 5xx, breaker, schema fail, or stale data. Every successful paid response carries an Ed25519-signed receipt your agent can verify. Open standard, machine-readable manifest at /.well-known/agent-fair-trade.json.

Read the standard

Payment Wallet

Address

0x549c82e6bfc54bdae9a2073744cbc2af5d1fc6d1View on Basescan

Network

Base mainnet

Currency

USDC (Circle native)

USDC contract

0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913

Cross-check this address before sending funds: /llms.txt, /api/payment/info, GitHub README, and the @tensorfeed bio. If any source disagrees, do not send.

Pricing

Base rate: 50 credits per $1 USDC (about $0.02 per credit). Volume discounts apply automatically when you buy larger bundles.

Send	Credits	Discount
$1.00	50	base
$5.00	275	10% off
$30.00	1,950	25% off
$200.00	16,000	40% off

Credits do not expire. Each premium call costs 1 to 5 credits depending on the tier. Tier 2 routing is currently 1 credit per call.

First-payment welcome bonus($1.00 of value)

The first successful USDC payment from a new sender wallet receives an extra 50 credits on top of the base rate. Granted automatically; the response from /api/payment/confirm includes welcome_bonus_credits and is_first_payment when applied. Same gate applies on the x402 fallback. Stacks with volume discounts: a $1 first payment yields 100 credits ($2.00 of value), a $5 first payment yields 325 credits.

Two Payment Flows

Credits-first (recommended)

Buy a batch of credits once, use a bearer token for all subsequent calls. About 50ms per call.

POST /api/payment/buy-credits with amount_usd and sender_wallet
Send USDC on Base FROM your sender_wallet to the returned wallet
POST /api/payment/confirm with tx_hash and the returned nonce
Receive a tf_live_* token; pass it as Authorization: Bearer

x402 fallback (one-off)

For agents that want to discover and pay in a single retry. About 3 to 4 seconds total latency on the first call. Requires a pre-flight quote so the on-chain sender can be bound and the public-mempool sniping vector is closed.

GET /api/premium/* with no auth
Receive 402 with payment instructions in headers and body
POST /api/payment/buy-credits with amount_usd and sender_wallet to bind a quote
Send USDC on Base FROM your sender_wallet to the publisher wallet
Retry with BOTH X-Payment-Tx and X-Payment-Quote (the returned memo) headers; receive data plus a token in X-Payment-Token header

Endpoints

GET/api/payment/infoFree

Public. Returns wallet address, pricing tiers, supported flows, and verification metadata.

Example response

{
  "ok": true,
  "wallet": {
    "address": "0x549c82e6bfc54bdae9a2073744cbc2af5d1fc6d1",
    "currency": "USDC",
    "network": "base",
    "contract": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913"
  },
  "pricing": { "base_rate": "50 credits per $1 USDC", "volume_bundles": [...] },
  "flow": { "with_quote": [...], "x402_fallback": [...] },
  "verification": { ... }
}

POST/api/payment/buy-creditsFree

Generate a 30-minute payment quote bound to a sender wallet. Returns wallet, memo (nonce), credit count, and the bound sender_wallet. Required body fields: amount_usd, sender_wallet.

Example response

// Body: { "amount_usd": 1.00, "sender_wallet": "0xYOUR_EOA" }
{
  "ok": true,
  "wallet": "0x549c82e6bfc54bdae9a2073744cbc2af5d1fc6d1",
  "sender_wallet": "0xyour_eoa",
  "memo": "tf-abc123",
  "amount_usd": 1.00,
  "credits": 50,
  "currency": "USDC",
  "network": "base",
  "expires_at": "2026-04-27T22:30:00Z",
  "ttl_seconds": 1800
}

POST/api/payment/confirmFree

Verify a USDC tx on Base and mint a bearer token. Requires nonce (legacy no-nonce path closed to prevent public-mempool tx sniping). The on-chain sender must match the quote-bound sender_wallet or the call returns sender_mismatch. Idempotent: same tx submitted twice is rejected.

Example response

// Body: { "tx_hash": "0x...", "nonce": "tf-abc123" }
// Note: nonce is REQUIRED. The on-chain tx.from MUST match the
// sender_wallet that was bound to the quote, or this returns
// { ok: false, error: "sender_mismatch" }.
{
  "ok": true,
  "token": "tf_live_<64-hex-chars>",
  "credits": 50,
  "balance": 50,
  "tx_amount_usd": 1.00,
  "rate": "base"
}

GET/api/payment/balanceToken required

Check remaining credits for the current bearer token.

Example response

// Header: Authorization: Bearer tf_live_...
{
  "ok": true,
  "balance": 47,
  "created": "2026-04-27T22:00:00Z",
  "last_used": "2026-04-27T22:14:23Z",
  "total_purchased": 50
}

GET/api/preview/routingFree (5/day/IP)

Free routing preview, top-1 model only, no score breakdown. Rate-limited to 5 calls per UTC day per IP.

Example response

{
  "ok": true,
  "preview": true,
  "task": "code",
  "rate_limit": { "limit": 5, "remaining": 4, "scope": "per IP per UTC day" },
  "recommendation": { "model": "Claude Opus 4.7", "provider": "anthropic" },
  "upgrade": { "premium_endpoint": "/api/premium/routing", ... }
}

GET/api/premium/routing1 credit per call

Tier 2 routing engine. Top-N ranked models with full composite score breakdown, pricing, status, and live data freshness. Custom weights via query params.

Example response

// Header: Authorization: Bearer tf_live_...
// Query: ?task=code&budget=5.0&top_n=3&w_quality=0.6&w_cost=0.3
{
  "ok": true,
  "task": "code",
  "weights": { "quality": 0.6, "availability": 0.0, "cost": 0.4, "latency": 0.0 },
  "recommendations": [
    {
      "rank": 1,
      "model": { "id": "claude-opus-4-7", "name": "Claude Opus 4.7", ... },
      "pricing": { "input": 15, "output": 75, "currency": "USD" },
      "status": "operational",
      "composite_score": 0.87,
      "components": { "quality": 0.94, "availability": 1.0, "cost": 0.65, "latency": 0.5 }
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 49 }
}

GET/api/premium/route-verdict1 credit per call

The signed model-routing decision. For ?task=code|reasoning|creative|general or ?model=<id-or-name>, returns the single best model to use right now, fusing pricing, benchmark capability discounted for contamination, real production usage, measured p95 latency probes, and live incident-triage operational state, plus runners-up and an AFTA-signed receipt over the inputs. Optional ?max_latency_p95_ms=, ?require_operational=, ?exclude_deprecated=. 30-minute operational freshness SLA, no-charge when the live layer is stale. A free top-verdict-only taste (no runners-up, no receipt) lives at /api/preview/route-verdict, 10 calls per IP per day.

Example response

// Header: Authorization: Bearer tf_live_...
// Query: ?task=code&max_latency_p95_ms=2000
{
  "ok": true,
  "query": { "task": "code", "model": null },
  "capturedAt": "2026-05-28T14:42:00Z",
  "verdict": {
    "rank": 1,
    "model": { "id": "claude-sonnet-4-6", "name": "Claude Sonnet 4.6", "provider": "anthropic", "openSource": false, "contextWindow": 200000 },
    "pricing": { "input": 3, "output": 15, "blended": 9, "currency": "USD", "unit": "per 1M tokens" },
    "quality": { "task_score": 0.87, "trust_discounted": 0.83, "contamination_note": "HumanEval (high contamination, saturated)" },
    "usage": { "corroborated": true, "rank": 1, "share_pct": 18.4, "trend": "flat" },
    "latency": { "measured_p95_ms": 1180, "source": "measured_probe" },
    "operational": { "ok": true, "status": "operational", "source": "live_status" },
    "deprecation": { "flagged": false, "status": null, "sunset_date": null },
    "composite_score": 0.78,
    "why": "code quality 0.83 after trust discount; corroborated by real usage (rank 1, 18.4% share, flat); measured p95 1180 ms; operational; blended $9 / 1M"
  },
  "runners_up": [
    { "rank": 2, "model": { "name": "GPT-5.5", "provider": "openai" }, "composite_score": 0.74 }
  ],
  "trust": { "usage_corroborated": true, "benchmark_contamination": "mixed", "operational_layer": "partial", "latency_layer": "partial" },
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

GET/api/premium/stack-safety-verdict1 credit per call

GO / HOLD / BLOCK deploy gate for your AI stack. Pass packages=name@version (up to 10) and get a verdict per package plus an overall gate, fusing the ingested AI-CVE batch (GHSA plus vendor advisories) with the CISA KEV catalog. Never-false-confirm: BLOCK only when an exploited CVE has no fix, HOLD when the pinned version must be verified, PASS when no AI-stack CVE matches the package name, UNKNOWN outside the curated cohort. A free gate-only preview (no CVE evidence, capped at 3 packages) lives at /api/preview/stack-safety-verdict, 10 calls per IP per day.

Example response

// Query: ?packages=langchain@0.3.27,vllm@0.6.0
{
  "ok": true,
  "gate": "HOLD",
  "capturedAt": "2026-05-27T09:00:00Z",
  "counts": { "block": 0, "hold": 1, "pass": 1, "unknown": 0 },
  "packages": [
    {
      "package": "vllm",
      "version": "0.6.0",
      "verdict": "HOLD",
      "in_cohort": true,
      "exploited": false,
      "fix_available": true,
      "category": "inference-stack",
      "matched_cves": [
        { "cve_id": "CVE-2026-1234", "on_kev": false, "exploited_in_wild": "stated_no", "severity_label": "high", "affected_version_ranges": ["< 0.6.1"], "fixed_versions": ["0.6.1"], "source_url": "https://github.com/advisories/GHSA-xxxx" }
      ],
      "reason": "A known CVE applies to this package (not confirmed exploited). Verify your pinned version against the surfaced ranges and fixes."
    },
    { "package": "langchain", "version": "0.3.27", "verdict": "PASS", "in_cohort": true, "exploited": false, "fix_available": false, "category": "agent-framework", "matched_cves": [], "reason": "No AI-stack CVE matched this package name. Not a full vulnerability scan." }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

GET/api/premium/benchmark-trust-verdict1 credit per call

Is an AI benchmark a trustworthy capability signal right now, or saturated, contaminated, or near its ceiling so a high score is a floor and not a differentiator? Returns a trust band and a 0-100 trust score per benchmark, fusing the registry contamination and saturation flags with the live spread of the top model scores (frontier compression), plus a down-weight recommendation and an alternative benchmark. Optional ?benchmark= or ?category=. A free band-and-score-only preview lives at /api/preview/benchmark-trust-verdict, 10 calls per IP per day.

Example response

// Query: ?category=code
{
  "ok": true,
  "capturedAt": "2026-05-28T08:00:00Z",
  "filter": { "benchmark": null, "category": "code" },
  "count": 7,
  "verdicts": [
    {
      "id": "humaneval", "name": "HumanEval", "category": "code", "status": "saturated", "contamination_risk": "high",
      "frontier_score": "~97%", "score_range": "0-100% pass@1", "trust_band": "contaminated", "trust_score": 12,
      "signals": { "ceiling_proximity": "at_ceiling", "frontier_compression": "unknown", "top_score_spread": null, "models_scored": 0 },
      "recommendation": "Down-weight scores on this benchmark (high training-contamination risk, marked saturated, frontier score is near the ceiling). A high score is closer to a capability floor than a differentiator. Prefer LiveCodeBench for current code signal.",
      "leaderboard_url": "https://paperswithcode.com/sota/code-generation-on-humaneval"
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

GET/api/premium/failover-verdict1 credit per call

Provider A is degraded, which operational provider do I fail over to for this task right now? Confirms A against the live incident-triage feed, then runs the capability-first route verdict with A and any provider flagged failover_now excluded, returning the recommended destination, the incident reason, and ranked alternatives. Param-required (?from=), optional ?task= or ?model=. A free destination-and-reason-only preview lives at /api/preview/failover-verdict, 10 calls per IP per day.

Example response

// Query: ?from=anthropic&task=code
{
  "ok": true,
  "from": { "provider": "anthropic", "in_incident": true, "incident": { "title": "Anthropic API elevated errors", "service": "Anthropic API", "impact_classification": "critical", "recommended_action": "failover_now", "started_at": "2026-05-28T11:00:00Z", "triage_summary": "Anthropic API returning elevated 5xx for Messages." } },
  "query": { "task": "code", "model": null },
  "capturedAt": "2026-05-28T11:55:00Z",
  "excluded_providers": ["anthropic"],
  "failover_to": { "rank": 1, "model": { "id": "gpt-5-5", "name": "GPT-5.5", "provider": "openai", "openSource": false, "contextWindow": 400000 }, "pricing": { "input": 5, "output": 15, "blended": 10, "currency": "USD", "unit": "per 1M tokens" }, "quality": { "task_score": 0.86, "trust_discounted": 0.83, "contamination_note": null }, "usage": { "corroborated": true, "rank": 2, "share_pct": 14.6, "trend": "up" }, "latency": { "measured_p95_ms": 1200, "source": "measured_probe" }, "operational": { "ok": true, "status": "operational", "source": "live_status" }, "deprecation": { "flagged": false, "status": null, "sunset_date": null }, "composite_score": 0.79, "why": "code quality 0.83 after trust discount; corroborated by real usage (rank 2, 14.6% share, up); measured p95 1200 ms; operational; blended $10 / 1M" },
  "alternatives": [],
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

GET/api/premium/sec/filings/guidance-delta1 credit per call

Did this periodic SEC filing (10-K or 10-Q) materially change guidance, segment outlook, or risk language versus the prior same-form filing, with the exact changed sentences quoted? Pass ?ticker=NVDA&form=10-Q for the latest same-form delta, or ?accession= for one specific filing. Returns a deterministic materiality_summary (counts by materiality, category, change type, and direction, plus a one-line headline) and the full changes array with verbatim prior and current quotes, values, and section. Input-keyed freshness: a filed 10-K or 10-Q is immutable, so the call no-charges only when a newer same-form filing has superseded the served delta. A free summary-only preview (quotes redacted) lives at /api/preview/sec/filings/guidance-delta, 10 calls per IP per day.

Example response

// Query: ?ticker=NVDA&form=10-Q
{
  "ok": true,
  "ticker": "NVDA",
  "company_name": "NVIDIA CORP",
  "form": "10-Q",
  "accession_number": "0001045810-26-000052",
  "prior_accession_number": "0001045810-25-000230",
  "filing_date": "2026-05-20",
  "prior_filing_date": "2025-11-19",
  "materiality_summary": { "total_changes": 10, "by_materiality": { "material": 3, "minor": 6, "boilerplate": 1 }, "by_category": { "revenue_guidance": 1, "segment_outlook": 2, "risk_factor": 4, "other": 3 }, "by_change_type": { "reworded": 8, "initiated": 1, "raised": 1 }, "by_direction": { "neutral": 8, "unclear": 1, "up": 1 }, "headline": "FY26 revenue guidance raised; Investments in Fiscal Year 2027 initiated; 4 risk factor wordings revised" },
  "changes": [ { "topic": "Investments in Fiscal Year 2027", "prior_text": "", "current_text": "In the first quarter of fiscal year 2027, we made the following investments: 18.6 billion dollars in private companies and infrastructure funds.", "prior_value": null, "current_value": "18.6 billion", "section": "Management's Discussion and Analysis", "category": "other", "change_type": "initiated", "direction": "unclear", "materiality": "material" } ],
  "freshness": { "model": "input_keyed", "superseded": false },
  "capturedAt": "2026-05-28T00:00:00Z",
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

GET/api/premium/provider-reliability-verdict1 credit per call

Which frontier AI provider is the most dependable to build on right now, and which is the riskiest? Ranks the providers TensorFeed actively probes by measured operational reliability, fusing availability (ok rate) and tail consistency (p50 over p95) from its own latency probes into one dependability score. The thesis: an agent retry loop feels the tail of the latency distribution, not the median, so the score rewards a tight tail alongside raw availability rather than crowning whoever wins the median. Returns the most-dependable and riskiest picks, the full per-provider ranking, and an AFTA-signed receipt. No params. 30-minute freshness SLA keyed to the probe computed_at, no-charge when stale. A free picks-only preview (no ranking, no receipt) lives at /api/preview/provider-reliability-verdict, 10 calls per IP per day.

Example response

// Header: Authorization: Bearer tf_live_...
{
  "ok": true,
  "capturedAt": "2026-05-29T11:55:00Z",
  "window_label": "last_24h",
  "verdict": { "most_dependable": "anthropic", "riskiest": "deepseek" },
  "ranking": [
    { "rank": 1, "provider": "anthropic", "ok_pct": 0.99, "total_p50_ms": 500, "total_p95_ms": 900, "total_p99_ms": 1400, "spread_ratio": 1.8, "reliability_score": 0.7728, "measured": true, "note": "measured availability 99 percent, p50 500 ms, p95 900 ms, p95 over p50 spread 1.8x" }
  ],
  "coverage": { "providers_ranked": 3, "fully_measured": 3, "availability_only": 0 },
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

GET/api/premium/x402-settlement-verdict1 credit per call

Is the Base x402 USDC settlement market growing, is it concentrated or a real market, and who leads? Rules over TensorFeed's own on-chain settlement index: momentum versus the prior window of equal length (expanding, steady, contracting, or nascent), concentration by the Herfindahl index (concentrated, moderate, or diversified), and the leading publisher, plus the full per-publisher volume ranking and the ecosystem totals with the window-over-window change. Optional ?window=24h|7d|30d (default 7d). Coverage is the Base settlements TensorFeed indexes, forward-only from launch: not all of x402 and not other chains. 10-minute freshness SLA keyed to the index cursor last_run_at, no-charge when stale. A free classifications-only preview lives at /api/preview/x402-settlement-verdict, 10 calls per IP per day.

Example response

// Query: ?window=7d
{
  "ok": true,
  "capturedAt": "2026-05-29T17:54:00Z",
  "window_label": "7d",
  "verdict": { "momentum": "expanding", "concentration": "concentrated", "leading_publisher": "pay.example.com" },
  "ecosystem": { "volume_usdc": "1234.500000", "count": 42, "unique_publishers": 3, "volume_change_pct": 30, "count_change_pct": 25, "prior_window_empty": false, "top_publisher_share_pct": 64.8, "hhi": 4908 },
  "ranking": [
    { "rank": 1, "domain": "pay.example.com", "volume_usdc": "800.000000", "count": 25, "share_pct": 64.8 }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 48 },
  "receipt": { "id": "rcpt_...", "signing_alg": "EdDSA", "signing_curve": "Ed25519", "signature": "<base64url>" }
}

POST/api/payment/trial-creditsFree

Free, zero-setup on-ramp. Sign an EIP-191 message proving you control a wallet (no on-chain transaction, no USDC, no gas), POST { message, signature }, and receive a bearer token preloaded with 25 trial credits. One grant per wallet, OFAC-screened, single-use nonce, 30-day expiry. Top up the same token later via /api/payment/buy-credits.

Example response

// POST body: { message, signature }. The wallet signs this EIP-191 message:
//   I am requesting TensorFeed trial credits for this wallet.
//
//   wallet: 0xYourEoa
//   timestamp: 2026-05-28T14:42:00.000Z
//   nonce: 9f3c1ab27de40581
{
  "ok": true,
  "token": "tf_live_<64-hex-chars>",
  "credits": 25,
  "expires_at": "2026-06-27T14:42:00.000Z",
  "wallet": "0xYourEoa",
  "how_to_use": "Send Authorization: Bearer <token> on any /api/premium/* call until the balance or the expiry runs out. Top up the same token via /api/payment/buy-credits."
}

GET/api/premium/history/pricing/series1 credit per call

Daily price points for one model across a date range, with min/max/delta summary and changes-detected count. Range capped at 90 days, default 30 days back.

Example response

// Query: ?model=Claude+Opus+4.7&from=2026-04-01&to=2026-04-27
{
  "ok": true,
  "model": "Claude Opus 4.7",
  "provider": "Anthropic",
  "range": { "from": "2026-04-01", "to": "2026-04-27", "days": 27 },
  "points": [
    { "date": "2026-04-01", "input": 18, "output": 90, "blended": 54 },
    { "date": "2026-04-15", "input": 15, "output": 75, "blended": 45 },
    { "date": "2026-04-27", "input": 12, "output": 60, "blended": 36 }
  ],
  "summary": {
    "first": { "date": "2026-04-01", "blended": 54 },
    "latest": { "date": "2026-04-27", "blended": 36 },
    "min_blended": 36, "max_blended": 54,
    "delta_pct_blended": -33.33,
    "changes_detected": 2,
    "days_with_data": 27, "days_missing": 0
  },
  "billing": { "credits_charged": 1, "credits_remaining": 48 }
}

GET/api/premium/history/benchmarks/series1 credit per call

Score evolution for a single benchmark on one model. Supported benchmark keys: swe_bench, mmlu_pro, gpqa_diamond, math, human_eval. Returns delta in percentage points.

Example response

// Query: ?model=Claude+Opus+4.7&benchmark=swe_bench&from=2026-04-01&to=2026-04-27
{
  "ok": true,
  "model": "Claude Opus 4.7",
  "benchmark": "swe_bench",
  "points": [
    { "date": "2026-04-01", "score": 70.0 },
    { "date": "2026-04-27", "score": 73.4 }
  ],
  "summary": { "min_score": 70.0, "max_score": 73.4, "delta_pp": 3.4 }
}

GET/api/premium/history/status/uptime1 credit per call

Daily status rollup for one provider over a date range. Returns operational/degraded/down day counts plus uptime % (degraded counts as half-credit). Missing-data days are excluded from the denominator.

Example response

// Query: ?provider=anthropic&from=2026-04-01&to=2026-04-27
{
  "ok": true,
  "provider": "anthropic",
  "days_total": 27, "days_with_data": 27, "days_missing": 0,
  "days_operational": 24, "days_degraded": 2, "days_down": 1, "days_unknown": 0,
  "uptime_pct": 92.59,
  "incident_days": [
    { "date": "2026-04-09", "status": "degraded" },
    { "date": "2026-04-17", "status": "down" }
  ]
}

GET/api/premium/history/news/full1 credit per call

Untruncated daily news archive. Single-date mode (?date=YYYY-MM-DD) returns the complete deduped article list for one UTC day (up to 200 articles). Range mode (?from=&to=, max 30 days) returns one entry per UTC date in the window. The free /api/history/news endpoint is capped at 25 articles per day; this endpoint exposes the full snapshot. Captured by every hourly RSS poll, so each day reflects the last poll before midnight.

Example response

// Query: ?date=2026-05-07
{
  "ok": true,
  "mode": "single",
  "date": "2026-05-07",
  "captured_at": "2026-05-07T23:00:14.812Z",
  "articles_count": 187,
  "articles": [
    {
      "id": "x9k2pq",
      "title": "Anthropic Ships Mythos to Defenders First",
      "url": "https://www.anthropic.com/news/mythos",
      "source": "Anthropic Blog",
      "sourceDomain": "anthropic.com",
      "snippet": "Anthropic released Mythos Preview today...",
      "categories": ["models", "security"],
      "publishedAt": "2026-05-07T18:30:00Z",
      "fetchedAt": "2026-05-07T19:00:11Z"
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 47 }
}

GET/api/premium/history/news/source-health1 credit per call

Multi-day per-source RSS poll reliability series. Each entry per day carries polls (total polls in the UTC day), polls_ok, polls_empty, polls_error, articles_total, reliability_pct, last_status, and last_error. Sources sorted by reliability_pct descending. Range capped at 90 days. Useful for trending source reliability, detecting feeds that went silent, or building a procurement-grade view of which AI publishers are most consistent.

Example response

// Query: ?from=2026-05-01&to=2026-05-07
{
  "ok": true,
  "from": "2026-05-01",
  "to": "2026-05-07",
  "days_returned": 7,
  "days": [
    {
      "date": "2026-05-07",
      "total_polls": 24,
      "updated_at": "2026-05-07T23:00:14.812Z",
      "sources": [
        { "id": "anthropic", "name": "Anthropic Blog", "polls": 24, "polls_ok": 24, "polls_empty": 0, "polls_error": 0, "articles_total": 32, "reliability_pct": 100, "last_status": "ok", "last_seen_at": "2026-05-07T23:00:11Z" },
        { "id": "huggingface", "name": "HuggingFace Blog", "polls": 24, "polls_ok": 18, "polls_empty": 6, "polls_error": 0, "articles_total": 14, "reliability_pct": 75, "last_status": "ok", "last_seen_at": "2026-05-07T23:00:12Z" }
      ]
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 46 }
}

GET/api/premium/security/cve/range1 credit per call

MITRE CVE List range query. Returns CVE IDs added or modified across a UTC date range, capped at 30 days. Each day in the response carries the full CVE-ID set indexed by the daily cvelistV5 commit-history scan. Pair with /api/security/cve/{id} for the per-CVE Record v5.2 lookup. License: MITRE CVE Terms of Use, commercial redistribution explicitly permitted; attribution block included on every response.

Example response

// Query: ?from=2026-05-06&to=2026-05-08
{
  "ok": true,
  "from": "2026-05-06",
  "to": "2026-05-08",
  "days_returned": 3,
  "cves_total": 412,
  "days": [
    { "date": "2026-05-06", "count": 138, "cve_ids": ["CVE-2026-1801", "CVE-2026-1802", "..."] },
    { "date": "2026-05-07", "count": 87,  "cve_ids": ["CVE-2026-1939", "CVE-2026-1940", "..."] },
    { "date": "2026-05-08", "count": 187, "cve_ids": ["CVE-2026-2027", "CVE-2026-2028", "..."] }
  ],
  "attribution": {
    "source": "MITRE CVE List",
    "source_url": "https://www.cve.org",
    "license": "MITRE CVE Terms of Use",
    "redistribution": "commercial-permitted"
  },
  "billing": { "credits_charged": 1, "credits_remaining": 45 }
}

GET/api/premium/cve/kev-exploitation-timeline1 credit per call

Per-vendor exploited-in-the-wild history from the cve-kev-2026 dataset. One vendor per call (?vendor=, loose matching: vendor=cisco resolves to "Cisco Systems, Inc."). Each CVE in the vendor timeline carries NVD disclosure date, days from disclosure to CISA KEV listing (often thousands of days for old CVEs CISA added in 2026), vulnerability class, CVSS, attack vector, vendor patch status, ransomware association, and a one-sentence summary, plus per-vendor aggregates (kev_count, mean and fastest KEV-add lag, severity distribution). Built from an offline per-CVE LLM extraction and a deterministic per-vendor rollup. v1 is a capped slice, not the full KEV catalog. License: NVD + CISA KEV US Government public domain (17 USC 105), commercial redistribution permitted.

Example response

// Query: ?vendor=cisco
{
  "ok": true,
  "vendor_query": "cisco",
  "matched_vendor": "Cisco Systems, Inc.",
  "dataset_meta": {
    "dataset": "cve-kev-2026",
    "cves_total": 200,
    "vendors_total": 91,
    "coverage": "v1 capped slice (not the full kev-anchored-2026 corpus)",
    "attribution": { "source": "NVD and CISA KEV", "license": "US Government public domain (17 USC 105)" }
  },
  "vendor": {
    "vendor_normalized": "Cisco Systems, Inc.",
    "cve_count": 12,
    "kev_count": 12,
    "ransomware_count": 0,
    "mean_days_disclosure_to_kev": 1180.4,
    "fastest_days_disclosure_to_kev": 21,
    "severity_distribution": { "critical": 5, "high": 7 },
    "timeline": [
      { "cve_id": "CVE-20XX-NNNN", "published_date": "20XX-0X-0X", "kev_date_added": "2026-0X-0X",
        "days_disclosure_to_kev": 412, "vulnerability_class": "rce-unauth", "cvss_v3_score": 9.8,
        "attack_vector": "network", "vendor_patch_status": "patch-released",
        "ransomware_use_known": false, "summary_one_sentence": "..." }
    ]
  },
  "billing": { "credits_charged": 1, "credits_remaining": 44 }
}

GET/api/premium/security/corroborated1 credit per call

One package whole corroborated GHSA advisory set per call (?package=, loose matching: package=commons-text resolves to the canonical name). Each advisory carries three explicit provenance buckets: corroborated_claim (verbatim affected_products + the deterministic product-vs-authoritative-OSV verdict, never-false-confirm), deterministic_enrichment (KEV/EPSS/SSVC/OSV joined ONLY by a verbatim-verified CVE id), verbatim_context (version/severity/exploited copied verbatim, not corroborated). meta.claim is surfaced verbatim: we corroborate the affected package and enrich by verified CVE id; we do NOT verify the advisory exploitation or severity claims (GHSA prose does not make them). Quarantined (extraction_suspect) records are never served. Honest counts: 82 package-addressable advisories (73 corroborated, 9 novel) across 47 packages; product-less unverifiable and quarantined excluded from served counts, not padded. Built from an offline grammar-constrained extraction, a deterministic OSV-only never-false-confirm corroboration, and a verbatim-CVE guard. Strict-premium: anonymous probes get the canonical 402. License: GHSA + OSV/CISA KEV/FIRST EPSS/NVD public; derived metadata + corroboration verdicts, advisory prose not republished.

Example response

// Query: ?package=free5gc
{
  "ok": true,
  "package_query": "free5gc",
  "matched_package": "free5GC",
  "claim": "Affected package corroborated against authoritative OSV, plus deterministic KEV/EPSS/CVSS/SSVC enrichment joined by a verbatim-verified CVE id. We do NOT verify the advisory exploitation or severity claims; GHSA prose does not make them.",
  "advisory_count": 13,
  "advisories": [
    {
      "source_url": "https://github.com/advisories/GHSA-27ph-8q4f-h7m7",
      "overall": "corroborated",
      "corroborated_claim": { "affected_products": ["free5GC"], "product_corroboration": "confirmed" },
      "deterministic_enrichment": {
        "cves_verbatim_verified": [], "kev_listed": false, "epss_percentile": null,
        "ssvc": null, "osv_packages": ["github.com/free5gc/bsf"]
      },
      "verbatim_context": {
        "affected_version_ranges": ["v4.2.1"], "fixed_versions": [],
        "severity_label": "unstated", "exploited_in_wild": "unstated"
      }
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 44 }
}

GET/api/premium/security/kev/full1 credit per call

Full untruncated CISA KEV catalog. Free /api/security/kev returns top-50 most-recent; this endpoint returns the entire current catalog with every field per entry. License: US Government public domain (17 USC 105), commercial redistribution explicitly permitted.

Example response

{
  "ok": true,
  "catalog_version": "2026.05.08",
  "date_released": "2026-05-08T17:31:07Z",
  "total_entries": 1590,
  "vulnerabilities": [
    {
      "cveID": "CVE-2026-42208",
      "vendorProject": "BerriAI",
      "product": "LiteLLM",
      "vulnerabilityName": "BerriAI LiteLLM SQL Injection",
      "dateAdded": "2026-05-08",
      "shortDescription": "...",
      "requiredAction": "Apply mitigations.",
      "dueDate": "2026-05-11",
      "knownRansomwareCampaignUse": "Unknown",
      "cwes": ["CWE-89"]
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 44 }
}

GET/api/premium/security/kev/series1 credit per call

Multi-day series of CISA KEV catalog additions across a UTC date range, capped at 90 days. Each day returns the full entry list whose dateAdded fell on that day. Useful for trending exploitation velocity, building anomaly detectors, or pulling weekly digest reports.

Example response

// Query: ?from=2026-05-01&to=2026-05-08
{
  "ok": true,
  "from": "2026-05-01",
  "to": "2026-05-08",
  "days_returned": 8,
  "total_added_in_range": 14,
  "days": [
    { "date": "2026-05-01", "count": 0, "entries": [] },
    { "date": "2026-05-07", "count": 3, "entries": [...] },
    { "date": "2026-05-08", "count": 11, "entries": [...] }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 43 }
}

GET/api/premium/security/epss/series1 credit per call

Full historical EPSS time-series for one CVE, sourced from FIRST.org. Each entry per date carries epss probability (0 to 1) and percentile rank. The series compounds with time as EPSS publishes daily. License: FIRST.org free-for-any-use policy.

Example response

// Query: ?cve_id=CVE-2024-3094
{
  "ok": true,
  "cve_id": "CVE-2024-3094",
  "score": {
    "cve": "CVE-2024-3094",
    "epss": "0.850580000",
    "percentile": "0.993590000",
    "date": "2026-05-08",
    "time-series": [
      { "epss": "0.850580000", "percentile": "0.993590000", "date": "2026-05-07" },
      { "epss": "0.84588000",  "percentile": "0.99337000",  "date": "2026-05-06" }
    ]
  },
  "billing": { "credits_charged": 1, "credits_remaining": 42 }
}

GET/api/premium/security/epss/top1 credit per call

Top-N highest-EPSS CVEs, optionally as of any historical UTC date. Free /api/security/epss/top serves the current snapshot only; this endpoint adds the historical-date filter. License: FIRST.org free-for-any-use policy.

Example response

// Query: ?date=2026-04-01&limit=5
{
  "ok": true,
  "date": "2026-04-01",
  "count": 5,
  "top": [
    { "cve": "CVE-2023-23752", "epss": "0.943100000", "percentile": "1.000000000", "date": "2026-04-01" }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 41 }
}

GET/api/premium/clean/cve/{CVE-id}1 credit per call

LLM-ready MITRE CVE record. Drops a typical CVE from ~3KB nested JSON to ~500 bytes flat JSON (~80% token reduction) with zero information loss for agent decision-making. Adds derived severity_band (none/low/medium/high/critical), deduped CWEs, flat vendor+product affected_products list, and top 5 references. Versioned via schema_version + cleaning_version for schema stability. The deep moat over the free /api/security/cve/{id} endpoint: agents pay because their context-window-tax savings exceed the $0.02 cost.

Example response

// GET /api/premium/clean/cve/CVE-2024-3094
{
  "ok": true,
  "source_format": "mitre_cve_v5_2",
  "target_format": "tensorfeed_llm_ready_v1",
  "source_payload": "cache",
  "schema_version": "1.0",
  "cleaning_version": "1.0",
  "transformed_at": "2026-05-09T03:45:11Z",
  "source": "MITRE CVE List",
  "data": {
    "id": "CVE-2024-3094",
    "state": "PUBLISHED",
    "published_at": "2024-03-29T16:51:12.588Z",
    "summary": "Malicious code in xz upstream tarballs.",
    "cvss_v3_1_score": 10,
    "cvss_v3_1_severity": "CRITICAL",
    "severity_band": "critical",
    "cwes": ["CWE-506"],
    "affected_products": ["xz liblzma", "Red Hat Fedora"],
    "affected_count": 2,
    "references_count": 8,
    "references_top": ["https://www.openwall.com/...", "https://access.redhat.com/...", "..."]
  },
  "billing": { "credits_charged": 1, "credits_remaining": 38 }
}

GET/api/premium/clean/kev/{CVE-id}1 credit per call

LLM-ready CISA KEV entry. Same content as /api/security/kev/{CVE-id} (free) but with normalized ransomware_use enum (yes/unknown/no) and notes_urls extracted from the upstream semicolon-separated reference text. Returns 404 if the CVE is not on the KEV catalog.

Example response

{
  "ok": true,
  "source_format": "cisa_kev_v1",
  "target_format": "tensorfeed_llm_ready_v1",
  "schema_version": "1.0",
  "cleaning_version": "1.0",
  "data": {
    "cve_id": "CVE-2026-42208",
    "vendor": "BerriAI",
    "product": "LiteLLM",
    "vulnerability_name": "BerriAI LiteLLM SQL Injection",
    "date_added": "2026-05-08",
    "due_date": "2026-05-11",
    "ransomware_use": "unknown",
    "cwes": ["CWE-89"],
    "notes_urls": ["https://github.com/BerriAI/litellm/security/advisories/...", "https://nvd.nist.gov/vuln/detail/..."]
  },
  "billing": { "credits_charged": 1, "credits_remaining": 37 }
}

GET/api/premium/clean/epss/{CVE-id}1 credit per call

LLM-ready EPSS score. Numeric probability (not stringified), derived risk_band (low/medium/high/critical), optional series summary (?series=true returns first/min/max snapshot instead of the full daily series, sized to fit comfortably in any context window). License: FIRST.org free-for-any-use policy.

Example response

// GET /api/premium/clean/epss/CVE-2024-3094?series=true
{
  "ok": true,
  "source_format": "first_org_epss_v1",
  "target_format": "tensorfeed_llm_ready_v1",
  "included_series": true,
  "schema_version": "1.0",
  "cleaning_version": "1.0",
  "data": {
    "cve_id": "CVE-2024-3094",
    "date": "2026-05-08",
    "epss_probability": 0.85058,
    "percentile": 0.99359,
    "risk_band": "high",
    "series_points": 365,
    "series_first": { "date": "2025-05-09", "epss": 0.71245 },
    "series_min": { "date": "2025-05-09", "epss": 0.71245 },
    "series_max": { "date": "2026-04-12", "epss": 0.86103 }
  },
  "billing": { "credits_charged": 1, "credits_remaining": 36 }
}

GET/api/premium/history/news/clusters/full1 credit per call

Full untruncated cross-source story clusters per UTC date. Each cluster groups TensorFeed news articles about the same story via embedding-based similarity and reports source_count, sources_list, hero article, corroboration_band (single / limited / broad), and contributing article_ids. The free /api/history/news/clusters caps at 25 clusters/day; this endpoint removes the cap and adds 30-day range support. Computed nightly at 07:30 UTC.

Example response

// GET /api/premium/history/news/clusters/full?date=2026-05-08
{
  "ok": true,
  "mode": "single",
  "date": "2026-05-08",
  "count": 47,
  "clusters": [
    {
      "cluster_id": "k3mn8q",
      "date": "2026-05-08",
      "article_count": 6,
      "source_count": 5,
      "sources": ["anthropic.com", "techcrunch.com", "theverge.com", "reuters.com", "bloomberg.com"],
      "article_ids": ["a1", "a2", "a3", "a4", "a5", "a6"],
      "hero": { "id": "a1", "title": "Anthropic Ships Mythos to Defenders First", "url": "https://www.anthropic.com/news/mythos", "source": "Anthropic Blog", "publishedAt": "2026-05-07T18:30:00Z" },
      "first_seen_at": "2026-05-07T18:30:00Z",
      "corroboration_band": "broad"
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 32 }
}

GET/api/premium/history/news/verified1 credit per call

Story-level feed filtered to clusters with N+ independent sources corroborating. Default min_sources=4 (broad corroboration). Agents asking "do not act on a single source" get a clean stream of stories that cleared the trust threshold. The verification product uniquely possible for TF because only TF has the cross-source view at scale.

Example response

// GET /api/premium/history/news/verified?from=2026-05-01&to=2026-05-08&min_sources=4
{
  "ok": true,
  "mode": "range",
  "from": "2026-05-01",
  "to": "2026-05-08",
  "min_sources": 4,
  "days_returned": 8,
  "total_verified": 23,
  "days": [
    {
      "date": "2026-05-08",
      "total_clusters_for_day": 47,
      "verified_count": 4,
      "clusters": [
        { "cluster_id": "k3mn8q", "source_count": 5, "corroboration_band": "broad", "hero": { "title": "Anthropic Ships Mythos..." } }
      ]
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 31 }
}

GET/api/premium/clean/fda/{category}1 credit per call

LLM-ready OpenFDA query results. Same five categories as the free /api/health/fda/{category} (drug/events, drug/labels, drug/recalls, food/recalls, device/events) with per-category flat schemas. Drug events flatten patient demo + drugs + reactions + outcomes + seriousness flags. Drug labels surface brand/generic/manufacturer + section text. Recalls expose classification + reason + voluntary flag. Device events extract primary device + outcomes + truncated narrative. License: CC0 1.0 Universal Dedication.

Example response

// GET /api/premium/clean/fda/drug/events?search=patient.drug.medicinalproduct:aspirin&limit=2
{
  "ok": true,
  "source_format": "openfda_v1",
  "target_format": "tensorfeed_llm_ready_v1",
  "schema_version": "1.0",
  "cleaning_version": "1.0",
  "data": {
    "category": "drug/events",
    "upstream_total": 609465,
    "count": 2,
    "results": [
      {
        "id": "10003304",
        "country": "US",
        "received_at": "2014-03-12",
        "serious": true,
        "seriousness_flags": ["hospitalization"],
        "patient_age": 65,
        "patient_sex": "female",
        "drugs": ["ASPIRIN"],
        "reactions": ["NAUSEA", "HEADACHE"],
        "primary_drug": "ASPIRIN",
        "primary_reaction": "NAUSEA",
        "drug_count": 1,
        "reaction_count": 2
      }
    ]
  },
  "billing": { "credits_charged": 1, "credits_remaining": 33 }
}

GET/api/premium/status/leaderboard1 credit per call

Cross-provider uptime ranking. Computed from minute-resolution counters (one sample every 2 minutes per provider, ~720 samples per provider per day). Each entry includes uptime_pct, polls, operational/degraded/down/unknown buckets, downtime_minutes, hard_down_minutes (excludes degraded), incident_count, and mttr_minutes (mean time to recover from resolved incidents). Sorted by uptime % DESC with hard_down_minutes as tie-breaker. Custom date range up to 90 days. Aimed at SRE/ops/procurement teams comparing AI vendor reliability.

Example response

// Query: ?from=2026-04-01&to=2026-04-30
{
  "ok": true,
  "range": { "from": "2026-04-01", "to": "2026-04-30", "days": 30 },
  "generated_at": "2026-05-04T21:00:00Z",
  "entry_count": 10,
  "poll_interval_minutes": 2,
  "entries": [
    {
      "provider": "Claude API",
      "rank": 1,
      "uptime_pct": 99.9722,
      "polls": 21600,
      "operational_polls": 21594,
      "degraded_polls": 6,
      "down_polls": 0,
      "unknown_polls": 0,
      "downtime_minutes": 12,
      "hard_down_minutes": 0,
      "incident_count": 1,
      "mttr_minutes": 12
    },
    {
      "provider": "OpenAI API",
      "rank": 2,
      "uptime_pct": 99.4444,
      "polls": 21600,
      "operational_polls": 21480,
      "degraded_polls": 100,
      "down_polls": 20,
      "unknown_polls": 0,
      "downtime_minutes": 240,
      "hard_down_minutes": 40,
      "incident_count": 4,
      "mttr_minutes": 35.5
    }
  ]
}

GET/api/premium/whats-new1 credit per call

Agent morning brief: pricing changes, new/removed models, status incidents, and top news headlines from the last 1-7 days. The single endpoint to call when an agent boots up.

Example response

// Query: ?days=1&news_limit=10
{
  "ok": true,
  "window": { "from": "2026-04-26T...", "to": "2026-04-27T...", "days": 1 },
  "summary": {
    "total_pricing_changes": 2, "new_models": 1, "removed_models": 0,
    "incidents": 1, "news_articles": 8
  },
  "pricing": {
    "changes": [
      { "model": "Claude Opus 4.7", "field": "inputPrice", "from": 15, "to": 12, "delta_pct": -20 }
    ],
    "new_models": [
      { "model": "Gemini 3.5", "provider": "Google", "input_per_1m": 6, "output_per_1m": 18, "tier": "flagship" }
    ],
    "removed_models": []
  },
  "status": {
    "incidents": [{ "service": "OpenAI", "severity": "minor", "title": "Degraded latency", "duration_minutes": 35 }],
    "currently_operational": 8, "currently_degraded": 0, "currently_down": 0
  },
  "news": [{ "title": "Anthropic ships ...", "url": "...", "published_at": "..." }],
  "billing": { "credits_charged": 1, "credits_remaining": 42 }
}

GET/api/premium/whats-new/pro10 credits per call

Pro tier of the morning brief. Same base payload as /api/premium/whats-new plus a Claude Haiku 4.5 analyst summary, 1 to 5 key takeaways with per-field cited basis IDs and confidence scores, and 1 to 3 recommended actions targeted by agent class. Every claim cites by stable basis ID assigned server-side BEFORE the model sees the data; citations are server-side validated, so the agent never sees a hallucinated reference. Failure to synthesize is a no-charge event under AFTA.

Example response

// Query: ?days=1&news_limit=10
{
  "ok": true,
  "tier": "pro",
  "window": { "from": "2026-05-13", "to": "2026-05-14", "days": 1 },
  "summary": { "total_pricing_changes": 1, "new_models": 1, "incidents": 1, "news_articles": 1 },
  "pricing": { "changes": [{ "model": "Claude Opus 4.7", "from": 15, "to": 14, "delta_pct": -6.6667 }], "new_models": [{ "model": "Sonnet 4.7", "tier": "mid" }] },
  "status": { "incidents": [{ "provider": "openai", "severity": "minor", "title": "Elevated latency" }] },
  "news": [{ "title": "Anthropic announces Sonnet 4.7", "url": "..." }],
  "data_ids": {
    "pricing_changes": { "c1": "Claude Opus 4.7|anthropic|inputPrice" },
    "new_models": { "m1": "Sonnet 4.7|anthropic" },
    "incidents": { "i1": "openai|Elevated latency|..." },
    "news": { "n1": "https://anthropic.com/news/sonnet-4-7" }
  },
  "pro": {
    "generated_by": "claude-haiku-4-5-20251001",
    "analyst_summary": "Anthropic cut Claude Opus 4.7 input pricing by 7 percent. New Sonnet 4.7 mid tier model announced. OpenAI had a minor 90 minute latency event...",
    "key_takeaways": [
      { "claim": "Anthropic cut Claude Opus 4.7 input pricing by 7 percent", "basis": ["c1"], "confidence": 0.98 },
      { "claim": "Anthropic announced a new Sonnet 4.7 mid tier model", "basis": ["m1", "n1"], "confidence": 0.95 }
    ],
    "recommended_actions": [
      { "for": "cost-bound", "action": "Re-evaluate Claude Opus as primary model given the price cut", "priority": "monitor", "basis": ["c1"] },
      { "for": "inference-bound", "action": "Test Sonnet 4.7 in low risk workloads", "priority": "monitor", "basis": ["m1"] }
    ]
  },
  "billing": { "credits_charged": 10, "credits_remaining": 32 }
}

GET/api/premium/probe/series1 credit per call

Daily SLA series for one LLM provider, measured by TensorFeed itself. Returns per-day count, success rate, ttfb p50/p95/p99, total latency p50/p95/p99, incident-hour count, plus an overall window summary. The free /api/probe/latest serves the most-recent 24h aggregate. We probe Anthropic, OpenAI, Google, Mistral, and Cohere chat-completion endpoints every 15 min with a single-token prompt and record measured response time + status. The data is unique: provider status pages are politically managed; this is what we measure. Pairs naturally with /api/premium/routing for picking a model whose SLA you can verify. 90-day max range, default 30 days back.

Example response

// Query: ?provider=anthropic&from=2026-04-01&to=2026-04-29
{
  "ok": true,
  "provider": "anthropic",
  "from": "2026-04-01",
  "to": "2026-04-29",
  "days": 29,
  "points": [
    {
      "date": "2026-04-29",
      "count": 96, "ok_pct": 0.989, "uptime_pct": 0.989,
      "ttfb_p50": 320, "ttfb_p95": 510, "ttfb_p99": 740,
      "total_p50": 410, "total_p95": 690, "total_p99": 1120,
      "incident_hours": 0, "has_data": true
    }
  ],
  "summary": { "overall_uptime_pct": 0.992, "days_with_data": 29, "days_with_incidents": 2 },
  "notes": [],
  "billing": { "credits_charged": 1, "credits_remaining": 41 }
}

GET/api/premium/openrouter/series1 credit per call

Daily OpenRouter cross-provider catalog drift over a 90-day window. Per day: model count, cheapest paid input and output USD-per-million floor, free-tier count, namespace breadth. Plus day-over-day churn: models added, models removed, and per-model price-change counts versus the prior captured day, with sample id lists. OpenRouter's public catalog serves only current state, so this longitudinal record is captured by TensorFeed and cannot be backfilled. Optional ?from=&to= (ISO dates), default 30-day window, 90-day max.

Example response

// Query: ?from=2026-05-01&to=2026-05-16
{
  "ok": true,
  "from": "2026-05-01",
  "to": "2026-05-16",
  "days": 16,
  "points": [
    {
      "date": "2026-05-16",
      "total_models": 372,
      "cheapest_input_usd_per_m": 0.02,
      "cheapest_output_usd_per_m": 0.05,
      "free_tier_count": 41,
      "namespace_count": 58,
      "top_namespace": "openai",
      "added": 3, "removed": 1, "price_changes": 7,
      "added_sample": ["x-ai/grok-4-mini"],
      "removed_sample": ["deprecated/old-model"],
      "has_data": true
    }
  ],
  "delta_in_window": {
    "start_total": 361, "end_total": 372, "net": 11,
    "cheapest_input_start": 0.03, "cheapest_input_end": 0.02,
    "cheapest_output_start": 0.06, "cheapest_output_end": 0.05
  },
  "notes": [],
  "billing": { "credits_charged": 1, "credits_remaining": 40 }
}

GET/api/premium/x402-registry/series1 credit per call

Daily x402 publisher-registry drift over a 90-day window. Per day: reachable vs erroring publisher counts, federation count, distinct payment networks, paid and free endpoint totals, and agent-fair-trade declarations. Plus day-over-day churn versus the prior captured day: domains added, domains removed, status flips (a publisher going dark or returning), and payment-wallet changes, each with sample domain lists. A registry is current-state only by nature, so this longitudinal record is captured by TensorFeed and cannot be backfilled. Optional ?from=&to= (ISO dates), default 30-day window, 90-day max.

Example response

// Query: ?from=2026-05-01&to=2026-05-16
{
  "ok": true,
  "from": "2026-05-01",
  "to": "2026-05-16",
  "days": 16,
  "points": [
    {
      "date": "2026-05-16",
      "total": 6, "ok_count": 5, "error_count": 1,
      "federation_count": 2, "network_count": 2,
      "networks": ["eip155:8453", "eip155:84532"],
      "paid_endpoints_total": 41, "free_endpoints_total": 12,
      "agent_fair_trade_count": 3,
      "added": 1, "removed": 0, "status_flips": 1, "wallet_changes": 0,
      "added_sample": ["newpublisher.ai"],
      "removed_sample": [],
      "wallet_change_sample": [],
      "has_data": true
    }
  ],
  "delta_in_window": {
    "start_total": 4, "end_total": 6, "net": 2,
    "start_ok": 4, "end_ok": 5
  },
  "notes": [],
  "billing": { "credits_charged": 1, "credits_remaining": 39 }
}

GET/api/premium/hf/velocity1 credit per call

Daily Hugging Face download-velocity over a 90-day window. Per day: the top models and datasets by download delta and the top Spaces by likes delta, computed only for ids present in the daily top-30 on both the prior captured day and this one. Plus top-set churn (models and datasets entered and exited) and window gainers (last minus first captured day). HF exposes only cumulative totals and a live top list, so this velocity is computed by TensorFeed and cannot be backfilled. Optional ?from=&to= (ISO dates), default 30-day window, 90-day max.

Example response

// Query: ?from=2026-05-01&to=2026-05-16
{
  "ok": true,
  "from": "2026-05-01",
  "to": "2026-05-16",
  "days": 16,
  "points": [
    {
      "date": "2026-05-16",
      "model_count": 30, "dataset_count": 30, "space_count": 30,
      "models_entered": 2, "models_exited": 2,
      "datasets_entered": 1, "datasets_exited": 1,
      "top_models_by_download_delta": [
        { "id": "deepseek-ai/DeepSeek-V4", "downloads": 4120333, "download_delta": 88210 }
      ],
      "top_datasets_by_download_delta": [
        { "id": "open-r1/Mixture", "downloads": 990122, "download_delta": 15400 }
      ],
      "top_spaces_by_likes_delta": [
        { "id": "huggingface/chat", "likes": 71233, "likes_delta": 412 }
      ],
      "has_data": true
    }
  ],
  "window": {
    "top_model_gainers": [
      { "id": "deepseek-ai/DeepSeek-V4", "downloads": 4120333, "download_delta": 612944 }
    ],
    "top_dataset_gainers": [
      { "id": "open-r1/Mixture", "downloads": 990122, "download_delta": 88010 }
    ],
    "model_count_start": 30,
    "model_count_end": 30
  },
  "notes": [],
  "billing": { "credits_charged": 1, "credits_remaining": 38 }
}

GET/api/premium/compare/models1 credit per call

Side-by-side comparison of 2-5 AI models. Each entry returns pricing, benchmarks (normalized to a union of keys with null for missing scores so downstream code never crashes on undefined), provider live status, capabilities, context window, and recent news. Plus rankings: cheapest blended, most context, and a per-benchmark leaderboard.

Example response

// Query: ?ids=opus-4-7,gpt-5-5,gemini-3
{
  "ok": true,
  "benchmark_keys": ["mmlu_pro", "swe_bench"],
  "models": [
    {
      "matched": true, "id": "opus-4-7", "name": "Claude Opus 4.7",
      "provider": "Anthropic", "tier": "flagship",
      "pricing": { "input": 15, "output": 75, "blended": 45 },
      "context_window": 1000000, "status": "operational",
      "benchmarks": { "swe_bench": 73.4, "mmlu_pro": 88.5 }
    }
  ],
  "rankings": {
    "cheapest_blended": [
      { "name": "Gemini 3", "blended": 14 },
      { "name": "GPT-5.5", "blended": 20 },
      { "name": "Claude Opus 4.7", "blended": 45 }
    ],
    "most_context": [
      { "name": "Gemini 3", "context_window": 2000000 }
    ],
    "by_benchmark": {
      "swe_bench": [
        { "name": "Claude Opus 4.7", "score": 73.4 },
        { "name": "GPT-5.5", "score": 70.0 }
      ]
    }
  },
  "billing": { "credits_charged": 1, "credits_remaining": 43 }
}

GET/api/premium/providers/{name}1 credit per call

Everything about an AI provider in one paid call: live status + components, all models with pricing + tier + benchmark scores joined, recent news mentions, agent traffic. Aggregation over 4 free endpoints in one paid request.

Example response

// Path: /api/premium/providers/anthropic
{
  "ok": true,
  "provider": { "id": "anthropic", "name": "Anthropic", "url": "https://anthropic.com" },
  "status": {
    "state": "operational",
    "status_page_url": "https://status.anthropic.com",
    "last_checked": "2026-04-27T18:00:00Z",
    "components": [{ "name": "API", "status": "operational" }]
  },
  "models": [
    {
      "id": "opus-4-7", "name": "Claude Opus 4.7", "tier": "flagship",
      "pricing": { "input": 15, "output": 75, "blended": 45 },
      "context_window": 1000000, "released": "2026-04-17",
      "capabilities": ["tool-use", "vision"],
      "benchmark_scores": { "swe_bench": 73.4, "mmlu_pro": 88.5 }
    }
  ],
  "recent_news": [{ "title": "Anthropic ships...", "url": "...", "published_at": "..." }],
  "recent_news_count": 12,
  "agent_traffic_24h": 124,
  "billing": { "credits_charged": 1, "credits_remaining": 44 }
}

GET/api/premium/cost/projection1 credit per call

Project the cost of a token-usage workload across 1-10 AI models. Returns daily/weekly/monthly/yearly totals per model plus a ranking by cheapest monthly. Pure compute on live /api/models pricing.

Example response

// Query: ?model=opus-4-7,gpt-5-5,gemini-3&input_tokens_per_day=1000000&output_tokens_per_day=500000&horizon=monthly
{
  "ok": true,
  "workload": { "input_tokens_per_day": 1000000, "output_tokens_per_day": 500000, "total_tokens_per_day": 1500000 },
  "primary_horizon": "monthly",
  "projections": [
    {
      "model": "Claude Opus 4.7", "provider": "Anthropic",
      "matched": true,
      "rates": { "input_per_1m": 15, "output_per_1m": 75, "blended_per_1m": 45 },
      "daily": { "input_cost": 15, "output_cost": 37.5, "total": 52.5 },
      "weekly_total": 367.5, "monthly_total": 1575, "yearly_total": 19162.5
    }
  ],
  "ranked_cheapest_monthly": [
    { "model": "Gemini 3", "provider": "Google", "monthly_total": 525 },
    { "model": "GPT-5.5", "provider": "OpenAI", "monthly_total": 750 },
    { "model": "Claude Opus 4.7", "provider": "Anthropic", "monthly_total": 1575 }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 46 }
}

GET/api/premium/news/search1 credit per call

Full-text search over the TensorFeed news corpus with relevance scoring (term hits in title weighted 3, snippet weighted 1, plus recency boost). Filter by date range, provider, and category. Omit q to browse the latest filtered articles. Default limit=25, max 100.

Example response

// Query: ?q=opus+pricing&provider=anthropic&from=2026-04-01
{
  "ok": true,
  "query": "opus pricing",
  "filters": { "provider": "anthropic", "from": "2026-04-01" },
  "total_corpus": 4823,
  "matched": 12,
  "returned": 12,
  "results": [
    {
      "title": "Anthropic ships Claude Opus 4.7 with 1M context",
      "url": "https://anthropic.com/opus-4-7",
      "source": "Anthropic Blog",
      "source_domain": "anthropic.com",
      "snippet": "New flagship model with extended context window...",
      "categories": ["Anthropic", "Models"],
      "published_at": "2026-04-27T12:00:00Z",
      "relevance": 0.85,
      "matched_terms": ["opus"]
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 47 }
}

GET/api/premium/agents/directory1 credit per call

The agents catalog joined with live status, recent news (count + top 3 articles), agent traffic, flagship pricing, and a derived trending_score (0-100). Server-side filter and sort so you pull a ranked list in one call. Default limit=50, max 100.

Example response

// Query: ?sort=trending&category=coding&status=operational
{
  "ok": true,
  "total": 18, "returned": 5,
  "sort": "trending",
  "filters_applied": { "category": "coding", "status": "operational" },
  "agents": [
    {
      "id": "claude-code", "name": "Claude Code", "provider": "Anthropic",
      "live_status": "operational",
      "status_page_url": "https://status.anthropic.com",
      "recent_news_count": 7,
      "recent_news": [
        { "title": "Anthropic ships Opus 4.7", "url": "...", "published_at": "..." }
      ],
      "agent_traffic_24h": 124,
      "flagship_pricing": { "model": "Claude Opus 4.7", "input": 15, "output": 75, "blended": 45 },
      "trending_score": 86
    }
  ],
  "billing": { "credits_charged": 1, "credits_remaining": 48 }
}

POST/api/premium/watches1 credit per registration

Register a webhook watch. Four types: realtime price (fires when a model price crosses a threshold or any change), realtime status (fires on provider operational/degraded/down transitions), scheduled digest (fires daily or weekly with a curated summary of pricing changes regardless of whether anything dramatic happened), or leaderboard_rank (fires when a provider crosses a rank threshold on the cross-provider 7-day uptime leaderboard). Each watch lives 90 days and fires up to 100 times by default. Deliveries POST to your callback URL with an HMAC-SHA256 signature header (X-TensorFeed-Signature: sha256=...). Per-token cap of 25 active watches. Listing and per-watch read/delete are free for the owning bearer token.

Example response

// Body: { spec, callback_url, secret?, fire_cap? }
// Spec types:
//   { type: "price", model, field: "inputPrice"|"outputPrice"|"blended",
//     op: "lt"|"gt"|"changes", threshold? }
//   { type: "status", provider, op: "becomes"|"changes",
//     value?: "operational"|"degraded"|"down" }
//   { type: "digest", cadence: "daily"|"weekly" }
//   { type: "leaderboard_rank", provider, op: "drops_below"|"rises_above"|"changes",
//     threshold? }  // threshold = integer rank (1=best); required unless op=changes
{
  "ok": true,
  "watch": {
    "id": "wat_a1b2c3d4e5f60718a9b0c1d2",
    "spec": { "type": "digest", "cadence": "daily" },
    "callback_url": "https://agent.example.com/hook",
    "created": "2026-04-27T18:00:00Z",
    "expires_at": "2026-07-26T18:00:00Z",
    "fire_count": 0, "fire_cap": 100, "status": "active"
  },
  "billing": { "credits_charged": 1, "credits_remaining": 49 }
}

// Digest fire payload (delivered to callback_url):
{
  "event": "watch.fire",
  "watch_id": "wat_...",
  "fired_at": "2026-04-28T07:00:00Z",
  "spec": { "type": "digest", "cadence": "daily" },
  "match": {
    "type": "digest",
    "cadence": "daily",
    "period": { "from": "2026-04-27T07:00:00Z", "to": "2026-04-28T07:00:00Z" },
    "pricing": {
      "changed": [
        { "model": "Claude Opus 4.7", "provider": "Anthropic",
          "field": "inputPrice", "from": 15, "to": 12, "delta_pct": -20 }
      ],
      "added": [], "removed": [], "total_changes": 1
    }
  }
}

GET/api/premium/research/milestones1 credit per call

Last 30 days of arXiv preprints flagged is_milestone_candidate by an offline Qwen 3.6 27B per-paper extraction. Each paper carries arxiv_id, date, subfield_tag, methodology_bucket, title, normalized affiliations, milestone_reasoning (named benchmark plus quantified delta, model release, or novel architecture), and a one-sentence summary. Conservative by design: false positives are worse than false negatives. Refreshed weekly. arXiv data is CC-BY; the per-paper extraction and milestone classification are the gate.

Example response

{
  "ok": true,
  "capturedAt": "2026-05-10",
  "window_days": 30,
  "total": 12,
  "papers": [
    {
      "arxiv_id": "2026.04001",
      "date": "2026-04-15",
      "subfield_tag": "llm-alignment",
      "methodology_bucket": "empirical-study",
      "title": "Constitutional AI for Multi-Turn Dialogue Safety",
      "affiliations": ["Anthropic"],
      "milestone_reasoning": "Established new SOTA on HHH-Eval at 87.4%, prior 81.2%.",
      "summary": "Extended constitutional AI methodology to multi-turn dialogue with new SOTA on HHH-Eval."
    }
  ],
  "attribution": { "source": "arXiv (preprint metadata) + TensorFeed Qwen-extracted analytical fields", ... },
  "billing": { "credits_charged": 1, "credits_remaining": 41 }
}

GET/api/premium/research/emerging-keywords1 credit per call

Top-50 multi-word keyphrases across recent arXiv abstracts, ranked by recent-vs-baseline lift. lift = (last 30d frequency) / (prior 90d frequency, smoothed). Each entry carries 2 to 5 example arxiv_ids so the agent can dive into a specific paper. Captures emerging research terminology before it shows up in citation counts. Refreshed weekly from the offline pipeline.

Example response

{
  "ok": true,
  "capturedAt": "2026-05-10",
  "recent_window_days": 30,
  "baseline_window_days": 90,
  "total": 50,
  "keywords": [
    {
      "keyword": "speculative decoding",
      "recent_count": 18,
      "baseline_count": 6,
      "lift": 9.1,
      "example_arxiv_ids": ["2026.04111", "2026.04222", "2026.04333"]
    }
  ],
  "attribution": { ... },
  "billing": { "credits_charged": 1, "credits_remaining": 40 }
}

GET/api/premium/funding/exposure1 credit per call

Derived analytics over the free /api/funding/portfolio registry. Returns silicon_concentration (per-silicon commitment count, total $ committed, share % of total), circular_exposure per investor (with loop_classification: fully-circular >=85%, partial-loop 25-85%, agnostic below, insufficient-data for one-shot investors), top_recipients sorted by inbound capital with the investor list per recipient, and co_investor_pairs (pairs of investors that both hold stakes in the same recipient). Underlying registry is editorial; the analytical layer is the gate.

Example response

{
  "ok": true,
  "capturedAt": "2026-05-10",
  "total_commitments": 8,
  "total_amount_usd_max": 313300000000,
  "silicon_concentration": [
    { "silicon_dependency": "tpu", "commitment_count": 1, "total_amount_usd_max": 200000000000, "share_of_total_pct": 63.84 },
    { "silicon_dependency": "nvidia", "commitment_count": 5, "total_amount_usd_max": 48300000000, "share_of_total_pct": 15.42 }
  ],
  "circular_exposure": [
    {
      "investor": "Nvidia",
      "investor_silicon_brand": "nvidia",
      "total_commitments": 3,
      "total_amount_usd_max": 35300000000,
      "commitments_to_own_silicon": 3,
      "circular_ratio_by_count": 1,
      "circular_ratio_by_amount": 1,
      "loop_classification": "fully-circular"
    }
  ],
  "top_recipients": [
    { "recipient": "OpenAI", "inbound_commitments": 2, "inbound_amount_usd_max": 43000000000, "investors": ["Microsoft", "Nvidia"] }
  ],
  "co_investor_pairs": [
    { "investor_a": "Microsoft", "investor_b": "Nvidia", "shared_recipients": ["OpenAI"] }
  ],
  "attribution": { ... },
  "billing": { "credits_charged": 1, "credits_remaining": 40 }
}

GET/api/premium/ai-companies/{ticker}1 credit per call

Per-ticker AI intelligence envelope for the 14 AI bellwethers (NVDA, AMD, AVGO, TSM, ARM, MSFT, GOOGL, AMZN, ORCL, PLTR, SMCI, AAPL, META, TSLA). Composes four free siblings into one paid call: latest 10 SEC filings (data.sec.gov, public domain), latest 10 AI-tagged news mentions filtered by curated aliases (so PLTR does not match unrelated 'palantir' matches), strategic and equity funding rounds where the company is a lead or notable investor in the TensorFeed funding registry, and cohort metadata (CIK, display name, category, AI angle, exchange). Single captured-at timestamp, 9h freshness SLA. Strict-premium path; ticker must be in the cohort, otherwise 404. Built for trading agents on Robinhood Agentic, ChatGPT, Claude, and custom rails doing pre-trade context in one round trip.

Example response

// GET /api/premium/ai-companies/NVDA
{
  "ok": true,
  "capturedAt": "2026-05-27T18:00:00Z",
  "ticker": "NVDA",
  "cohort_size": 14,
  "company": {
    "ticker": "NVDA",
    "cik": "0001045810",
    "display_name": "NVIDIA",
    "category": "silicon",
    "ai_angle": "The dominant supplier of AI training and inference GPUs (H100, H200, B200, Rubin). Every frontier lab buys from here.",
    "exchange": "NASDAQ"
  },
  "filings": {
    "count": 10,
    "items": [
      {
        "accession_number": "0001045810-26-000052",
        "form": "10-Q",
        "filing_date": "2026-05-20",
        "primary_doc_url": "https://www.sec.gov/Archives/edgar/data/1045810/..."
      }
    ],
    "source": "data.sec.gov/submissions",
    "license": "Public domain (17 USC 105). SEC EDGAR."
  },
  "news": {
    "count": 4,
    "items": [
      {
        "id": "abc123",
        "title": "NVIDIA announces Rubin successor at GTC",
        "url": "https://example.com/nvda-rubin",
        "source": "TechCrunch AI",
        "publishedAt": "2026-05-27T14:00:00Z",
        "matched_aliases": ["NVIDIA"]
      }
    ],
    "aliases_used": ["NVIDIA", "Nvidia"],
    "sanitization": "enabled"
  },
  "funding_as_investor": {
    "count": 0,
    "items": [],
    "description": "Strategic and equity rounds where this company is listed as a lead or notable investor in TensorFeed funding registry."
  },
  "attribution": { "sources": [ "SEC EDGAR ...", "TensorFeed news ...", "TensorFeed funding ..." ] },
  "billing": { "credits_charged": 1, "credits_remaining": 40 }
}

GET/api/premium/research/topic-search1 credit per call

Structured search over the arXiv preprint corpus using the TF derived taxonomy. Filters: subfield_tag, methodology_bucket, since (YYYY-MM-DD), until (YYYY-MM-DD), milestone_only (1 or 0), limit (1 to 100, default 25), offset (paginate). subfield_tag is one of llm-architecture, llm-training, llm-eval, llm-alignment, multimodal, vision, speech, robotics, rl, theory, optimization, efficiency, retrieval, agents, code-generation, scientific-ml, fairness-safety, dataset, survey, application, other. methodology_bucket is one of new-architecture, training-recipe, fine-tuning, eval-benchmark, theoretical-analysis, empirical-study, dataset-release, system-tooling, survey, position-paper, application, other. Returns matching papers sorted by date desc. arXiv's native search has no concept of methodology bucket or our subfield taxonomy; that's the gate.

Example response

// Query: ?subfield_tag=agents&methodology_bucket=system-tooling&since=2026-04-01&limit=5
{
  "ok": true,
  "capturedAt": "2026-05-10",
  "query": {
    "subfield_tag": "agents",
    "methodology_bucket": "system-tooling",
    "since": "2026-04-01",
    "until": null,
    "milestone_only": false,
    "limit": 5,
    "offset": 0
  },
  "total_matches": 73,
  "returned": 5,
  "papers": [
    {
      "arxiv_id": "2026.05003",
      "date": "2026-05-05",
      "title": "Agent Memory via Retrieval Over Episodic Logs",
      "subfield_tag": "agents",
      "methodology_bucket": "system-tooling",
      "is_milestone_candidate": false,
      "affiliations": ["MIT", "OpenAI"],
      "summary": "Introduced an episodic-log retrieval mechanism for long-running agent memory."
    }
  ],
  "attribution": { ... },
  "billing": { "credits_charged": 1, "credits_remaining": 39 }
}

GET/api/premium/research/lab-productivity1 credit per call

Top labs by paper count over rolling 30d / 90d / 365d windows on the offline Qwen-extracted arXiv corpus. Each entry carries affiliation, papers (count), and type (industry, academia, government, nonprofit, mixed, unknown). Filters: window (30d|90d|365d, default returns all three), affiliation_type (one of the type values), limit per window (1 to 50, default 25). arXiv has no native concept of normalized lab attribution; the per-paper extraction and affiliation normalization are the gate.

Example response

// Query: ?window=90d&affiliation_type=industry&limit=5
{
  "ok": true,
  "capturedAt": "2026-05-10",
  "query": {
    "window": "90d",
    "affiliation_type": "industry",
    "limit": 5
  },
  "windows": {
    "90d": [
      { "affiliation": "Google DeepMind", "papers": 142, "type": "industry" },
      { "affiliation": "Meta AI", "papers": 98, "type": "industry" },
      { "affiliation": "Microsoft Research", "papers": 81, "type": "industry" },
      { "affiliation": "OpenAI", "papers": 47, "type": "industry" },
      { "affiliation": "Anthropic", "papers": 31, "type": "industry" }
    ]
  },
  "attribution": { ... },
  "billing": { "credits_charged": 1, "credits_remaining": 38 }
}

GET/api/premium/x402-index/publisher/{domain}1 credit per call

Per-publisher receipt feed for one x402-compliant domain across an inclusive from/to date range. Returns publisher meta (domain, pay_to_wallets, first_seen) plus the window rollup (volume_usdc, count, avg_amount, daily_series of per-day rows for the entire range, zeros included). Forensic and compliance lane for any caller building dashboards over x402 settlement data on Base mainnet. Required params: from, to (YYYY-MM-DD). Strict-premium path; anonymous Bazaar probes see a clean 402 challenge before the param check. Returns 404 with a hint when the domain is not present in the x402 publisher registry (see /api/x402-index/publishers for the indexed set). Wave 20 Bazaar pilot.

Example response

// GET /api/premium/x402-index/publisher/tensorfeed.ai?from=2026-05-21&to=2026-05-27
{
  "ok": true,
  "publisher": {
    "domain": "tensorfeed.ai",
    "pay_to_wallets": ["0x549c82e6bfc54bdae9a2073744cbc2af5d1fc6d1"],
    "first_seen": "2026-05-28T00:00:00Z"
  },
  "window": { "from": "2026-05-21", "to": "2026-05-27", "days": 7 },
  "rollup": {
    "volume_usdc": "287.420000",
    "count": 1342,
    "avg_amount": "0.214172",
    "daily_series": [
      { "date": "2026-05-21", "volume_usdc": "32.180000", "count": 156 },
      { "date": "2026-05-22", "volume_usdc": "41.260000", "count": 198 },
      { "date": "2026-05-23", "volume_usdc": "38.940000", "count": 182 },
      { "date": "2026-05-24", "volume_usdc": "44.180000", "count": 211 },
      { "date": "2026-05-25", "volume_usdc": "39.620000", "count": 188 },
      { "date": "2026-05-26", "volume_usdc": "47.860000", "count": 219 },
      { "date": "2026-05-27", "volume_usdc": "43.380000", "count": 188 }
    ]
  },
  "attribution": "TensorFeed x402 settlement index over public Base mainnet on-chain data",
  "license": "CC BY 4.0",
  "billing": { "credits_charged": 1, "credits_remaining": 37 }
}

GET/api/premium/x402-index/series1 credit per call

Time-series of ecosystem-level or per-publisher x402 USDC settlement volume or count across a from/to date range. Chart-feeding shape. Required params: metric (volume | count), granularity (day | hour), from, to (YYYY-MM-DD). Optional domain filter (omit for ecosystem-level series). MVP supports granularity=day; granularity=hour returns an empty series with an attribution note that hour granularity is not yet available. Each series row is { ts, value }: value is a USDC decimal string when metric=volume, an integer when metric=count. Strict-premium path; anonymous Bazaar probes see a clean 402 challenge before the param check. Wave 20 Bazaar pilot.

Example response

// GET /api/premium/x402-index/series?metric=volume&granularity=day&from=2026-05-23&to=2026-05-27
{
  "ok": true,
  "metric": "volume",
  "granularity": "day",
  "window": { "from": "2026-05-23", "to": "2026-05-27" },
  "series": [
    { "ts": "2026-05-23", "value": "38.940000" },
    { "ts": "2026-05-24", "value": "52.180000" },
    { "ts": "2026-05-25", "value": "47.620000" },
    { "ts": "2026-05-26", "value": "58.860000" },
    { "ts": "2026-05-27", "value": "54.380000" }
  ],
  "attribution": "TensorFeed x402 settlement index over public Base mainnet on-chain data",
  "billing": { "credits_charged": 1, "credits_remaining": 36 }
}

Code Examples

Python SDKpip install tensorfeed

from tensorfeed import TensorFeed

tf = TensorFeed()

# 1. Quote a credit purchase (30-minute TTL). sender_wallet is REQUIRED.
quote = tf.buy_credits(amount_usd=1.00, sender_wallet="0xYOUR_EOA")
print(f"Send {quote['amount_usd']} USDC on Base FROM {quote['sender_wallet']}")
print(f"          TO {quote['wallet']}")
print(f"Memo: {quote['memo']}")

# 2. Send the USDC tx FROM the same sender_wallet you bound to the
#    quote (Rabby, Coinbase Wallet, etc.) then confirm with the tx
#    hash. The on-chain from must match sender_wallet exactly.
result = tf.confirm(tx_hash="0xYOUR_TX_HASH", nonce=quote["memo"])
# The token is auto-stored on the client; routing() uses it automatically.

# 3. Call premium endpoints
rec = tf.routing(task="code", budget=5.0, top_n=3)
for r in rec["recommendations"]:
    print(f"#{r['rank']}: {r['model']['name']} (score: {r['composite_score']:.2f})")

# 4. Check remaining credits
print(tf.balance())

Shellcurl, credits-first flow

# Free preview (5 calls/day per IP)
curl "https://tensorfeed.ai/api/preview/routing?task=code"

# 1. Quote a purchase. sender_wallet is REQUIRED and binds the quote
#    to the EOA that will sign the on-chain USDC transfer. This closes
#    the public-mempool tx sniping vector.
curl -X POST https://tensorfeed.ai/api/payment/buy-credits \
  -H "Content-Type: application/json" \
  -d '{"amount_usd": 1.00, "sender_wallet": "0xYOUR_EOA"}'

# 2. Send USDC on Base from sender_wallet to the wallet returned in
#    step 1, then confirm with the tx hash and nonce. Both fields are
#    REQUIRED; legacy no-nonce confirm path is closed.
curl -X POST https://tensorfeed.ai/api/payment/confirm \
  -H "Content-Type: application/json" \
  -d '{"tx_hash": "0xYOUR_TX", "nonce": "tf-abc123"}'

# 3. Use the token on premium endpoints
curl -H "Authorization: Bearer tf_live_..." \
  "https://tensorfeed.ai/api/premium/routing?task=code&top_n=5"

# 4. Check balance
curl -H "Authorization: Bearer tf_live_..." \
  https://tensorfeed.ai/api/payment/balance

Shellcurl, x402 fallback flow

# x402 flow with sender-wallet binding (post-2026-05-05)
curl https://tensorfeed.ai/api/premium/routing
# 402 Payment Required + payment instructions in headers and body

# 1. Pre-flight quote bound to your sender wallet
curl -X POST https://tensorfeed.ai/api/payment/buy-credits \
  -H "Content-Type: application/json" \
  -d '{"amount_usd": 1.00, "sender_wallet": "0xYOUR_EOA"}'
# Response includes "memo": "tf-abc123"

# 2. Send USDC on Base from sender_wallet to the publisher wallet.

# 3. Retry the premium endpoint with BOTH X-Payment-Tx and
#    X-Payment-Quote headers. Both are REQUIRED on the x402 fallback;
#    the worker checks the on-chain sender against the quote-bound
#    sender_wallet and rejects sender_mismatch otherwise.
curl -H "X-Payment-Tx: 0xYOUR_TX" \
     -H "X-Payment-Quote: tf-abc123" \
  "https://tensorfeed.ai/api/premium/routing?task=code"
# Returns the data + a fresh token in X-Payment-Token header for future calls

Circuit Breaker (Loop Protection)

Two independent layers run on every premium call. Both return HTTP 429 with no credits charged.

Identical-request layer (trip_kind: "identical_request"): more than 20 requests with the same path and sorted query string from one bearer token inside a 60s rolling window trips a 120s cooldown. The error code is infinite_loop_detected. Catches naive while-loops.
Burn-rate layer (trip_kind: "burn_rate"): more than 100 requests from one bearer token inside the same 60s window, regardless of path or query, trips a 300s cooldown. The error code is burn_rate_exceeded. Catches loops that vary the URL on each call (for example, appending a random nonce parameter) and would otherwise slip past the identical-request layer.

Different tokens are tracked independently, so normal high-volume traffic across multiple agents is unaffected. If you legitimately need more than 100 RPM on one token, split traffic across multiple tokens or email evan@tensorfeed.ai.

Signed Receipts & Agent Nonce

Every premium response (paid OR no-charge) carries an Ed25519-signed receipt your agent can verify against the public key at /.well-known/tensorfeed-receipt-key.json with no shared secret. The signature covers a canonical-JSON form of the receipt fields, so any tampering with the body, the credit charge, or the no-charge reason invalidates the signature.

Send an optional X-Agent-Nonce header with any premium request and the value is echoed verbatim into the signed receipt's agent_nonce field. This binds the signature to your specific request, so a server cannot return a previously signed receipt from a cached identical call. The nonce is also returned in the X-Agent-Nonce-Echo response header for log correlation.

Allowed characters: [A-Za-z0-9._-]. Length: 8 to 128 characters.
Recommended: a UUIDv4 (with dashes stripped) or a base64url-encoded random 32-byte value. Generate a fresh nonce on every call.
Receipts are versioned: v: 2 includes agent_nonce (null when no nonce was supplied); v: 1 receipts (issued before 2026-05-03) omit the field and verify under the v1 canonical form.

Verification reference: /api/receipt/verify. The full canonical-JSON spec is at /agent-fair-trade#receipts.

Rate Limits (Free Tier)

Free public endpoints (everything outside /api/premium/*) are rate-limited to 120 requests per minute per IP. Premium bearer tokens skip this limit entirely (they are gated by credits and the per-token circuit breaker instead).

Every response broadcasts standard headers so a well-behaved agent can pace itself without trial and error:

RateLimit-Limit total requests allowed per window.
RateLimit-Remaining requests left in the current window.
RateLimit-Reset seconds until the window resets.

Equivalent X-RateLimit-* headers are also sent for older clients. On a 429, Retry-After tells you exactly how long to wait.

Prompt-Injection Sanitization

Aggregated text on the agent endpoints (/api/news,/api/agents/news, /feed.xml, /feed.json) is scrubbed at read time before it reaches your agent. We strip ASCII control chars, bidi/zero-width spoofing chars, and neutralize role-confusion tokens like <|im_start|>, [INST], and lines that try to fake a system or assistant turn. Title and snippet are also length-capped so a single feed item cannot fill your context window.

URLs, source attribution, dates, and categories are passed through untouched. The endpoint advertises sanitization: "enabled" in the JSON response so you can verify the layer is on.

Chaos Engineering Headers (Free)

Test your fallback logic without waiting for a real upstream incident. Both headers are free, no-auth, work on every endpoint, and never charge credits.

X-TensorFeed-Simulate-Error: 503 returns the requested status code (any value in 400-599) with a body that declares the response is simulated.
X-TensorFeed-Simulate-Latency: 2500 sleeps the requested milliseconds before responding (capped at 10000ms).

curl -H "X-TensorFeed-Simulate-Error: 503" https://tensorfeed.ai/api/news
curl -H "X-TensorFeed-Simulate-Latency: 2500" https://tensorfeed.ai/api/status

Simulated responses include the header X-TensorFeed-Simulated: true so your test assertions can verify they are coming from the chaos layer and not a real outage.

Model Context Protocol (MCP)

TensorFeed ships two MCP transports for two distinct deployment shapes. Same tool surface, different connection model.

Hosted HTTP MCP (Streamable HTTP, MCP 2024-11-05)

The canonical entry for hosted-marketplace listings (Anthropic vertical agent repos, claude.ai connectors, third-party MCP catalogs). POST a JSON-RPC 2.0 envelope to https://mcp.tensorfeed.ai/mcp; GET returns discovery info. CORS open for cross-origin agent fetches. 12 tools in V1 spanning AI news, model pricing, AI service status, MITRE CVE, CISA KEV, EPSS, OSV.dev, SEC EDGAR (search + submissions + ticker lookup), and EIA Open Data. The legacy https://tensorfeed.ai/api/mcp path still works for backward compatibility.

# Probe + initialize
curl https://mcp.tensorfeed.ai/mcp

# Tool list
curl -X POST https://mcp.tensorfeed.ai/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# Tool call: ticker lookup
curl -X POST https://mcp.tensorfeed.ai/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"lookup_sec_company_ticker",
                 "arguments":{"ticker_or_cik":"AAPL"}}}'

To add as a Claude Code plugin or vertical-agent MCP entry, point the client's .mcp.json at the URL with type: "http". No API key required for V1 free tools.

npm stdio MCP server (client-side install)

For Claude Desktop, Claude Code CLI, Cursor, Cline, and any MCP client that prefers spawning a local subprocess. Same tool surface, plus access to premium tools when you set the TENSORFEED_TOKEN environment variable.

# One-shot install + run (no global install needed)
npx -y @tensorfeed/mcp-server

# Or, for premium tools, set the bearer token first:
export TENSORFEED_TOKEN=tf_live_...
npx -y @tensorfeed/mcp-server

Buy a bearer token via /account or programmatically via /api/payment/buy-credits + /api/payment/confirm. Both transports speak the same MCP spec; pick whichever matches how your agent connects.

Free Preview Tier

Every paid endpoint has a free preview at /api/preview/*. The free tier returns the top recommendation only (no score breakdown) and is rate-limited to 5 calls per UTC day per IP. The counter resets at UTC midnight.

Use the preview to validate the endpoint before committing credits. After 5 calls in a day, the response is a 429 with a hint to use the paid endpoint and the number of hours until reset.

Terms summary

No-training license (17.1): Premium API responses are licensed for inference use only. Use for training, fine-tuning, evaluation, or distillation of ML models is prohibited.
No refunds (17.5): All credit purchases are final and non-refundable. Credits do not expire and remain redeemable across tensorfeed.ai and terminalfeed.io. Buy small (for example, $1 USDC for 50 credits), calibrate, then top up as needed.
Best-effort, no SLA (17.6): We aim for high uptime but offer no service guarantee. Specific premium endpoints may be modified or discontinued with reasonable notice.
Replay protection (17.4): Each USDC tx can be used to mint credits exactly once. Re-submitting the same tx hash is rejected.
Cross-site bundle (17.8): Bearer tokens and credits are jointly redeemable on tensorfeed.ai and terminalfeed.io. TensorFeed is the system of record for credit balances.
Sanctions and jurisdiction (17.9): Premium API access is unavailable to OFAC-, EU-, UK-, or UN-sanctioned persons and to residents of comprehensively sanctioned jurisdictions (Cuba, Iran, North Korea, Syria, Crimea, Donetsk, Luhansk). Inbound credit-purchase transactions are screened against the Chainalysis public sanctions API.
Acceptable use (17.12): No reselling of bearer tokens, no proxy APIs that materially reproduce the Premium API surface, no use of premium responses to train or evaluate competing models.
Governing law and venue: California law, exclusive venue Los Angeles County, California.

This is a summary, not the full agreement. Full legal terms are in the Terms of Service. Premium API data practices are in the Privacy Policy.

Free API docs

News, status, models, history

Python SDK

pip install tensorfeed

/api/payment/info

Live wallet and pricing JSON