TensorFeed

Enriched AI agents directory. One paid call returns the catalog joined with live provider status, recent news mentions, 24h agent traffic, and sample model pricing, ranked by a derived trending score. The agent-discovers-agent feed.

AI-thesis token pulse: joins price moves with venue-weighted perp funding-rate skew. Per-token setup classification (squeeze_up = rising + negative funding = shorts trapped; chase_up = rising + positive funding = leverage mean-reversion risk; squeeze_down; chase_down; coiled = flat + extreme funding). Notable movers cohort + breadth + median change summary. Filters: token substring, setup kind, min_abs_change_pct.

AI-stack CVE intelligence. Filters fresh security disclosures to vendors and products in the AI inference, agent framework, training, vector DB, model gateway, and MCP tool layers. Each paper carries a tf_ai_category tag and a severity_rank, sorted with exploited-in-wild and critical-severity first. 99.8% of underlying data is from GitHub Security Advisories (CC BY 4.0).

AI-stack CVEs actively exploited in the wild. Returns only the AI inference, agent-framework, training, vector DB, and model-gateway vulnerabilities flagged as exploited (stated_yes), ranked by severity, with affected products, fixed versions, and the source advisory. The "which AI dependency is being weaponized right now and do I have a patch" call. 99.8% GitHub Security Advisories (CC BY 4.0).

AI safety incident exposure analytics. Per-developer and per-deployer incident counts with recency-weighted exposure_score (1.0 last 30d, 0.5 days 31-90, 0.25 older), risk_domain and SEP-view (Security/Ethics/Performance) distributions, top affected artifacts. Derived over the AVID (avidml/avid-db, MIT) snapshot refreshed daily. The "which AI vendors have the most recent reported safety incidents" call.

AI-package security radar. Per-package risk_score (0-100) over the daily OSV snapshot of curated AI/ML PyPI + npm packages: critical_count_30d * 25 + high_30d * 12 + critical_90d * 6 + high_90d * 3 + min(open_count, 20) + 5 if any advisory in last 7d, saturated to 100. Risk_band classifications (calm <10 / watch 10-25 / hot 25-50 / critical 50+) plus notable_movers (top-5 by_critical_30d, by_risk_score, new_in_last_7d). The "which AI deps should I be worried about right now" call.

AI-relevant APIs from the APIs.guru directory of 2400+ OpenAPI specs, filtered via curated keyword matching. Per-entry first_seen_at against TensorFeed snapshot history so an agent can answer "what new AI APIs appeared in the last 7 days": a diff APIs.guru itself cannot provide. CC-BY-SA 4.0; source links preserved.

Benchmark Trust Verdict. Is an AI benchmark a trustworthy capability signal right now, or is it 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 to use instead. Optional ?benchmark= or ?category= filter. AFTA-signed. The "should I trust this SOTA claim" call.

LLM-ready CVE record. Around 80 percent token reduction versus raw MITRE v5.2, with derived severity_band, deduped CWEs, flat affected_products list, and the top 5 references. Pass the CVE id as the URL path segment.

LLM-ready FIRST.org EPSS score with derived risk_band (none, low, moderate, elevated, high, critical). Pass the CVE id as the URL path segment. Optional series=true returns first/min/max summary instead of the full daily-score series.

LLM-ready CISA KEV entry with normalized ransomware_use enum (known, unknown, not_used) and notes_urls extracted from the freeform notes field. Pass the CVE id as the URL path segment. Returns 404 with a hint if the CVE id is valid but not on the KEV catalog.

Side-by-side comparison of 2-5 AI models. Returns pricing, benchmarks normalized to union-of-keys with explicit nulls, live provider status, recent news mentions, and per-benchmark rankings in one ready-to-rank payload.

Project the cost of a token-usage workload across 1-10 AI models. Returns daily, weekly, monthly, and yearly totals per model plus a ranking by cheapest monthly. The canonical "given my workload, what would each model cost me" call.

Failover Verdict. 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 model plus the incident reason plus ranked alternatives plus an AFTA-signed receipt. Pass ?from=<provider> and optional ?task= or ?model=. The "A just broke, where do I send it" call.

Cross-provider arbitrage analytics for hosted inference. Per-model: cheapest_paid, most_expensive_paid, spread_usd, savings_pct, median_paid_blended, fastest_tps, cheapest_with_tps, free_tier_offers. Plus provider_rollup (cheapest_count, top_tps_count, value_score 0-100) and top_arbitrage models sorted by savings_pct desc. Pairs with the free /api/inference-providers matrix. The "where to migrate workloads" call.

Model deprecation timeline with migration intelligence. Each entry enriched with urgency_band (within_7d / within_30d / within_60d / within_90d / past), days_until_sunset, days_since_sunset, and a resolved migration_chain hop sequence to the first still-active replacement. Optional within_days window centered on now; optional provider filter. The "is my model affected and what should I migrate to" call.

AI-package release velocity. Per-package releases_24h / releases_7d / releases_30d, latest bump classification (major/minor/patch/prerelease, pre-1.0 minor counts as major per semver), is_breaking_recent flag (major bump within 30 days). Notable movers: recent_major_bumps, most_releases_7d, fastest_cadence_30d. Filter by ecosystem (PyPI/npm), category (llm-sdk, agent-framework, etc), or package substring. The "what AI deps just changed and is it breaking" call.

Forward and backward calendar over the AI policy registry. Each entry classified relative to now (past / active / upcoming), days-until-effective, with next-3-milestones for upcoming items. Filter by jurisdiction (EU, US, UK, etc) and time window. The "what AI rules are about to bite" feed.

Provider Reliability Verdict. One paid call ranks the frontier AI providers TensorFeed actively probes by measured operational reliability, fusing availability (ok rate) and tail consistency (p50 over p95) from its own latency probes into a single dependability ranking. Returns the most dependable and riskiest providers, the full per-provider ranking (availability, p50/p95/p99, spread ratio, score), and an AFTA-signed receipt. The thesis: an agent retry loop feels the tail, not the median. The "which provider can I depend on right now" call.

Top 100 AI researchers ranked by AI publication volume in the trailing 365 days, enriched with h-index, i10-index, total citation count, primary affiliation (institution + country), ORCID, and derived AI-share-of-total. OpenAlex CC0. Daily refresh.

Top 100 recent AI papers ranked by the share of their total citations earned in the most recent calendar year. Papers published in the last 2 years with 3+ citations. Surfaces papers that are still gaining traction now. OpenAlex CC0.

Top 50 multi-word keyphrases across recent arXiv AI abstracts ranked by recent-vs-baseline lift (last 30 days frequency over prior 90 days, smoothed). Each keyphrase carries 2-5 example arxiv_ids so an agent can drill into the underlying work. The trend-spotter feed for AI research vocabulary.

Last 30 days of arXiv preprints flagged as milestone candidates by an offline Qwen 3.6 27B per-paper extraction pass. Each paper carries structured reasoning naming the benchmark + quantified delta, model release, or novel architecture justification. Conservative; false positives are worse than false negatives.

Per-institution AI research publication velocity. 30-day window vs 365-day baseline with direction classification, notable movers, plus by-country and by-affiliation-type breakdowns. OpenAlex CC0. The "which AI labs are accelerating" feed.

Signed model routing decision. One paid call returns the best model for a task (code, reasoning, creative, general) or a named model, fusing pricing, benchmark capability discounted for contamination, real production usage, measured p95 latency from active probes, live incident-triage operational state, and deprecation flags into a single verdict with runners-up and an AFTA-signed receipt over the inputs. Optional filters: max_latency_p95_ms, require_operational, exclude_deprecated. The "which model right now" call.

Top-N AI model recommendation for a task. Joins live pricing, benchmarks, provider status, and latency into a single composite score so an agent gets a ranked answer in one call instead of stitching three free endpoints together.

Cross-database CVE verification. Composes MITRE CVE plus CISA KEV plus FIRST.org EPSS plus OSV.dev plus CISA Vulnrichment into one fact card with a confirmed_by array and corroboration_count. The single-call anti-hallucination lookup for security agents. Pass the CVE id as the URL path segment.

Stack Safety Verdict. A GET deploy gate: pass packages=name@version,name@version (up to 10) from your AI stack and get GO / HOLD / BLOCK per package plus an overall gate, fusing TensorFeed's ingested AI-CVE batch (GHSA plus vendor advisories) with the CISA KEV catalog for exploitation status. Each verdict carries the matched CVE ids, affected ranges, fixed versions, and KEV status, plus an AFTA-signed receipt. Never-false-confirm: BLOCK only on exploited with no fix, HOLD when the version must be verified, PASS on no AI-stack match, UNKNOWN outside the cohort. The "is my AI stack safe to ship" call.

Agent morning brief. One paid call returns the last 24 hours of AI pricing changes, new and removed models, service-status incidents, and top news headlines. The endpoint an agent calls on boot.

x402 Settlement Verdict. One paid call rules on the state of the x402 USDC settlement market on Base, computed over TensorFeed's own on-chain settlement index. Returns the market momentum (expanding, steady, contracting, or nascent versus the prior window of equal length), the concentration by Herfindahl index (concentrated, moderate, or diversified), the leading publisher, the full per-publisher volume ranking, ecosystem totals with the window-over-window change, and an AFTA-signed receipt. Optional window (24h, 7d, 30d). The "is x402 settlement growing and who leads" call, over the Base settlements TensorFeed measures, not a market-wide claim.

Derived structural metrics over the AI funding portfolio. Silicon-vendor concentration shares per investor, circular-loop classification (fully-circular / partial-loop / agnostic), co-investor pairs holding the same recipient, top recipients by inbound capital. The "who is funding whose silicon" call.

AI/ML PyPI package momentum and velocity. Per-package week-over-week change, direction classification (accelerating / steady / declining), notable movers, and by-category counts. The "what AI packages are heating up" feed.

Per-provider digest. Pass the provider slug (anthropic, openai, google, etc.) as the URL path segment. Returns the model catalog scoped to that provider plus current pricing matrix, status incidents, and reliability summary. The single-call lookup an agent makes when it needs everything about a specific provider.