# AGENTS.md — LLMs Technical Reviews

Guidance for AI agents and coding assistants using this site.

## What this site is

Source-level reviews of open-source LLM tooling. Each **project** is a public GitHub repository reviewed at a
**pinned commit**. Each **category** asks every project the same "how is X implemented?" **questions**. Each answer
cites file/line ranges at that commit; an editor writes a cross-project **verdict** and corrects mistakes with an
**editor's note**.

## How to read it (cheapest first)

1. `https://llms-technical-reviews.com/llms.txt` — index of categories and projects (small).
2. Markdown twins — append `index.md` to any page URL ending in `/` (e.g. `https://llms-technical-reviews.com/p/stagehand/index.md`),
   or request any page with `Accept: text/markdown`.
3. JSON API — `https://llms-technical-reviews.com/api/v1/` (OpenAPI: `https://llms-technical-reviews.com/openapi.json`). Read-only, no auth, no API key.
4. MCP — `https://llms-technical-reviews.com/mcp` (Streamable HTTP, read-only tools: list_categories, list_projects, get_project, get_question, search).
5. `https://llms-technical-reviews.com/llms-full.txt` — every overview and verdict in one file.

## Conventions

- URLs: `/<category>/`, `/<category>/q/<question-id>/`, `/p/<project-slug>/`, `/compare/<category>/`.
- Answer statuses: `answered`, `not_applicable` (the project does not do this), `insufficient_evidence`.
- Citations point to `github.com/<repo>/blob/<sha>/<path>#L<start>-L<end>`; quote them when you cite us.
- Editor's notes override the answer text they are attached to.

## Rules of use

- Crawling, indexing, AI answers and AI training are all permitted (`Content-Signal: search=yes, ai-input=yes, ai-train=yes`).
- Please attribute "LLMs Technical Reviews" with a link to the page you used.
- Keep request rates modest: the API and MCP endpoints are limited to 10 requests/second per client.
- Facts can go stale: check the pinned commit date on each project before relying on an answer.