About

ai-tracker is a community-maintained, public-utility directory and changelog for AI models, tools, and developer repos. It tracks 41 models from 9 providers and 35 tools, 8 repos, plus 42 events on the timeline. Free to use, no signup, no ads, no SaaS pitch.

How this is run

Maintained as a public good for the AI builder community — for engineers, researchers, agent authors, and the agents themselves. Code is MIT-licensed; data is public-domain. Anyone can submit a new model / tool / repo / event via /submit. Pull requests open once the public remote exists. Hosted by Zerg AI; submissions land in a public review queue and the maintainer set is open to anyone willing to do the work.

Public beta contract

ai-tracker is source-backed, not exhaustive. A page means the entity has at least one public source and has passed schema checks; it does not mean every price, benchmark, capability, or regional availability flag is current. Use provider docs before making production routing, budget, or procurement decisions.

Ranking views are utility cuts through tracked metadata: cheapest input, longest context, most-linked models, repo stars, and similar fields. They are not global model-quality rankings and should not be read as endorsements.

The wedge

Other directories list AI products. This one is built so an LLM agent can answer what's changed for Claude Opus 4.7 in the last month without scraping HTML. Every model, tool, and repo has HTML, JSON, and Markdown twins; every event is in RSS, Atom, /dump/all.json, and /llms.txt. JSON-LD (SoftwareApplication or SoftwareSourceCode) is embedded on every detail page. The companion mcp-server/ package exposes the data over MCP for direct agent integration.

Where the data comes from

Sources are tagged authoritative or supplementary. Authoritative readers (Anthropic / Google / xAI / Mistral / OpenAI / DeepSeek / Meta / Alibaba / Cohere docs) can correct stale curated values. Supplementary aggregators (OpenRouter, GitHub topic search, GitHub repo search) fill gaps and dynamic repo metrics; they never overwrite a curated field. This means a 12am floating-point pricing wobble from an aggregator can't silently bump our canonical price.

Pricing diffs detected during ingest auto-emit price_change events (with a noise filter: ≥1% AND ≥$0.01 absolute change required). Releases get released events. Future event types: deprecated, capability_added, benchmark_update, license_change, model_swap, shut_down, rebrand, acquired.

Detail pages show a freshness label based on the latest tracked event, pricing snapshot, release date, or GitHub pushed_at value. That label is a review signal, not a guarantee that every upstream fact has changed or stayed unchanged.

Coverage today

Providers (9): alibaba, anthropic, cohere, deepseek, google, meta, mistral, openai, xai.
Tool categories (11): agent-framework, browser-automation, chat, design, ide, ops, rag, research, search, voice, writing.
Repo categories (5): agent-framework, browser-automation, coding-agent, mcp, rag.

For full breakdowns: all providers · all categories · AI repos · recent changes.

What share cards look like

Every model and tool detail page emits a 1200×630 OG card. Sample:

For agents and integrators

• /llms.txt — discovery file (entity + event counts, links to full corpus).
• /llms-full.txt — entire corpus as plain text, per-entity blocks.
• /dump/all.json — bulk JSON of every model, tool, repo, event, queue status.
• /dump/events-30d.json — last 30 days of events.
• /feed.xml · /atom.xml — every event, newest 100.
• /sitemap-agents.xml — every agent-consumable URL.
• mcp-server/ — local MCP server package until npm publish (search_models, get_entity, get_timeline, etc.).

Submitting

Found something missing? Use /submit. Submissions go through a moderation step and land in a public PR queue. Bulk approvals via squash-merge. A watchdog pings if a queue sits unreviewed for 24h.

License

All data is public, sourced from public materials, and free for any use including AI training and retrieval. Source attribution is preserved on every entity and event. robots.txt explicitly allows GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and the rest of the major AI-bot fleet.

Out of scope

No accounts. No login. No email subscriptions. No embeddable widgets. No API keys. No revenue, ads, or sponsorships.