tensorfeed/ai-ecosystem-daily

TensorFeed AI Ecosystem Daily

Daily snapshots of the AI ecosystem: news, model pricing, benchmarks, service status, GPU rental prices, MCP registry growth, LLM endpoint latency probes, agent traffic, and the AFTA adopter directory. Captured once per day from the public tensorfeed.ai API and committed to this repo as JSONL.

Each daily snapshot lives in a YYYY-MM-DD/ subfolder with one JSONL file per feed plus a manifest.json summarizing what was captured.

What's in here

File	Records	Description
news.jsonl	up to 200	AI news articles aggregated from major sources, snippets clipped, prompt-injection sanitized.
models.jsonl	~230	Model pricing and specs across all major labs, flattened with provider name on each row.
pricing.jsonl	1 (summary)	Compact pricing payload for agents.
status.jsonl	~12	Real-time operational status of major AI services.
benchmarks.jsonl	~5	Public benchmark scores per model.
agents-directory.jsonl	~18	Curated AI agent directory.
agents-activity.jsonl	varies	Live AI bot traffic on tensorfeed.ai (ClaudeBot, GPTBot, Applebot, etc).
podcasts.jsonl	~50	Recent AI podcast episodes.
trending-repos.jsonl	~20	Trending GitHub repos in AI/ML.
mcp-registry.jsonl	1 (summary)	Daily count + delta of the official MCP server registry.
probe.jsonl	1 (summary)	Last 24h of LLM endpoint latency measurements.
gpu-pricing.jsonl	1 (summary)	GPU rental price snapshot across cloud marketplaces (Vast.ai, RunPod).
afta-adopters.jsonl	varies	Sites publishing an AFTA manifest.
ai-hardware.jsonl	~17	AI accelerator specs: NVIDIA Hopper/Blackwell, AMD Instinct, Google TPU, AWS Trainium, Apple, Cerebras, Groq. FLOPS, VRAM, memory bandwidth, list price.
open-weights.jsonl	~9	Production-ready open-weights models with quantization options (FP16/FP8/AWQ INT4/GGUF), VRAM requirement per quantization, recommended GPU class, license.
inference-providers.jsonl	~8	Hosted inference providers for open-weights models (Together, Fireworks, etc) with per-model pricing.
training-runs.jsonl	~11	Disclosed and estimated training cost catalog: parameters, training tokens, hardware, GPU hours, USD millions, costSource (disclosed vs estimated).
marketplaces.jsonl	~12	AI marketplace catalog: GPT Store, Claude Skills, HF Spaces, HF Models, Replicate, MCP Registry, etc.
specialized-models.jsonl	~19	Domain-specialized models (code, medical, legal, finance, music, 3D, retrieval).
fine-tuning.jsonl	~12	Fine-tuning providers (first-party + hosted) with training pricing per 1M tokens, methods, base models supported.
oss-tools.jsonl	~25	Production OSS tools agents and developers actually install (Ollama, llama.cpp, vLLM, Open WebUI, etc).
agent-apis.jsonl	~29	Non-LLM APIs agents commonly wire (Tavily, Brave, Exa, Firecrawl, OpenWeather, Stripe, Twilio, etc).
voice-leaderboards.jsonl	1 (summary)	TTS Arena Elo + Open ASR Leaderboard WER rankings.
embeddings.jsonl	~18	Embedding + reranker model catalog.
multimodal.jsonl	~24	Image, video, TTS, STT model catalog with pricing.
vector-dbs.jsonl	~12	Vector database catalog (managed, OSS, hybrid).
frameworks.jsonl	~15	Agent framework catalog (LangChain, CrewAI, AG2, etc) with language + category.
benchmark-registry.jsonl	~24	Benchmark catalog with active/saturated status.
public-leaderboards.jsonl	~20	Pointers to every live public AI leaderboard.
conferences.jsonl	~18	AI conferences (research, industry, developer) with dates.
model-deprecations.jsonl	~12	Provider model retirement and deprecation calendar (OpenAI, Anthropic, Google, Cohere, etc). Each row: provider, model, status (announced / deprecated / sunsetted), announced/deprecation/sunset dates, recommended replacement model, source URL pointing to the provider's own announcement.
funding.jsonl	~21	AI funding rounds catalog by stage and category.
model-cards.jsonl	~8	Per-model system / safety / red-team document index.
ai-policy.jsonl	~10	AI regulation tracker (EU AI Act, US executive orders, etc).
compute-providers.jsonl	~17	GPU cloud, hyperscaler, AI-serverless, marketplace catalog.
usage-rankings.jsonl	~20	Model usage rankings (which models are actually winning).
agent-provisioning.jsonl	~18	Agent infrastructure providers (hosting, DB, auth, observability) by AFTA-style protocol status.
training-datasets.jsonl	~19	Pretraining + post-training dataset catalog (RedPajama, FineWeb, OpenAssistant, etc) with token counts, licenses, and stage.
mcp-servers.jsonl	~30	Curated MCP server catalog with capabilities, transport, and first-party flags. Distinct from mcp-registry.jsonl which captures the official-registry day-over-day count.
attention.jsonl	~12	Per-provider attention index derived from news, trending repos, and bot traffic. One row per provider per day for time-series analysis of mindshare.
incidents.jsonl	varies	Service incident history surfaced by TensorFeed status monitoring.
harnesses.jsonl	~4	Agentic-coding harness leaderboard snapshot (SWE-bench, Aider, etc) with vendor self-reported best scores per base model.
embodied-ai.jsonl	~25	Vision-language-action foundation models (pi-0, GR00T, OpenVLA), humanoid platforms (Figure 02, 1X NEO, Optimus), real-world training datasets (Open X-Embodiment, DROID), and physics simulators.
ai-lawsuits.jsonl	~20	Structured catalog of active AI litigation: NYT v OpenAI, Authors Guild, Getty, Concord, RIAA v Suno + Udio, Doe v GitHub, Andersen v Stability AI, Garcia v Character.AI, FTC + EU inquiries. Each row: parties, court, case number, claims, status, stage, primary-source citations. Editorial summaries; not legal advice.
x402-adopters.jsonl	~10	Catalog of publishers, SDKs, gateways, and references speaking the x402 HTTP-payment protocol. TensorFeed.ai, TerminalFeed.io (live), Stripe Link Agents (announced), @coinbase/x402 SDK, afta-gateway template, tensorfeed-mcp.

Quick start

The dataset is published in two formats. Pick the one that fits your stack.

Hugging Face datasets library (Python, JSONL on the main branch)

from datasets import load_dataset

# Latest news
news = load_dataset("tensorfeed/ai-ecosystem-daily", "news", split="train")

# Model pricing time series (load all dates, filter by date column or filename)
models = load_dataset("tensorfeed/ai-ecosystem-daily", "models", split="train")

# Model deprecation calendar
deprecations = load_dataset("tensorfeed/ai-ecosystem-daily", "model-deprecations", split="train")

DuckDB (SQL directly against the Parquet branch, no Python deps)

The same data is available as Parquet on the refs/convert/parquet branch, auto-generated by Hugging Face. DuckDB can query it directly with no install of datasets.

-- Latest deprecation announcements across all providers
SELECT provider, model, status, deprecation_date, replacement
FROM 'https://huggingface.co/datasets/tensorfeed/ai-ecosystem-daily/resolve/refs%2Fconvert%2Fparquet/model-deprecations/train/0000.parquet'
ORDER BY deprecation_date DESC;

Pandas (read Parquet directly)

import pandas as pd

url = "https://huggingface.co/datasets/tensorfeed/ai-ecosystem-daily/resolve/refs%2Fconvert%2Fparquet/news/train/0000.parquet"
df = pd.read_parquet(url)

ClickHouse (analytics workloads on the time-series feeds)

SELECT provider, AVG(latency_ms) AS avg_latency
FROM url('https://huggingface.co/datasets/tensorfeed/ai-ecosystem-daily/resolve/refs%2Fconvert%2Fparquet/probe/train/0000.parquet', Parquet)
GROUP BY provider
ORDER BY avg_latency;

Update cadence

Snapshots commit at 08:00 UTC each day, fired by a GitHub Actions workflow in the TensorFeed repo. One commit per day. The data itself is captured in the TensorFeed Worker via the daily 07:00 UTC captureHistory cron, so the file written here reflects state at roughly 07:00 UTC.

Cannot be backfilled. Every day without a snapshot is a day of history lost.

License: inference-only

This dataset is released under TensorFeed's inference-only license. You may use it as input context for AI agents and LLM inference (RAG, evals, prompt context, agent tools). You may not use it as training data for foundation models without explicit written permission.

The full Agent Fair-Trade Agreement (AFTA) v1.0 is published at tensorfeed.ai/whitepaper and the public-facing summary at tensorfeed.ai/agent-fair-trade. Both documents cover the rationale, the receipt-signing infrastructure, the federation pattern, and the machine-payable upgrade path for the live API tier.

Live API

Everything in this dataset is also available as a live API. If you need fresher than daily, prefer the API:

• Free, no-auth: tensorfeed.ai/developers
• Premium tier (USDC on Base): tensorfeed.ai/developers/agent-payments
• OpenAPI 3.1 spec: tensorfeed.ai/openapi.yaml
• MCP server: npm install -g @tensorfeed/mcp-server

Citation

@misc{tensorfeed_ai_ecosystem_daily,
  title  = {TensorFeed AI Ecosystem Daily},
  author = {{TensorFeed.ai}},
  year   = {2026},
  publisher = {Hugging Face},
  url    = {https://huggingface.co/datasets/tensorfeed/ai-ecosystem-daily}
}

Built with Claude (Anthropic). The full system, including the AFTA standard governing this dataset's license, was designed in collaboration with Claude. Source: github.com/RipperMercs/tensorfeed.