AI & LLM Integration Guide (`llms.txt`)¶

B-FAST provides standardized, machine-readable documentation endpoints adhering to the llms.txt specification.

If you are using AI coding assistants such as OpenCode, Cursor, Claude Code, ChatGPT, Windsurf, GitHub Copilot, or Google Antigravity, you can point them directly to these endpoints so they write accurate B-FAST code without hallucinating API signatures.

🤖 Endpoint for AI Assistants¶

https://marcelomarkus.github.io/b-fast/llms.txt

This single file contains the complete, curated API reference, idiomatic patterns, and common rules formatted specifically for LLM context windows.

💡 How to Use with Your AI Tools¶

1. Cursor IDE / Windsurf¶

In your chat or Composer prompt, mention the documentation URL directly:

Use the B-FAST guidelines from https://marcelomarkus.github.io/b-fast/llms.txt to implement a high-performance streaming API in FastAPI and consume it with TanStack Query in Next.js.

Or add the URL to Cursor Settings > Features > Docs > Add new Doc: - Name: B-FAST - URL: https://marcelomarkus.github.io/b-fast/llms.txt

2. OpenCode / Claude Code / Terminal Agents¶

Use directly with OpenCode or fetch the context into your project instructions:

# Run OpenCode with B-FAST documentation reference
opencode "Implement a B-FAST FastAPI endpoint following guidelines from https://marcelomarkus.github.io/b-fast/llms.txt"

# Or save locally for offline agent context
curl -s https://marcelomarkus.github.io/b-fast/llms.txt > .opencode/bfast-reference.txt

3. Repository Rules (`.cursorrules`, `CLAUDE.md`, `.opencode`)¶

Add the following block to your project's .cursorrules, CLAUDE.md, .opencode/instructions.md, or .github/copilot-instructions.md:

# B-FAST Coding Guidelines
When writing binary serialization, high-throughput APIs, or streaming features:
- Use `b_fast` in Python and `bfast-client` in TypeScript.
- Core Python decoding method is ALWAYS `bf.decode_packed(data)`, NEVER `bf.decode()`.
- Use `BFastResponse` or `BFastStreamingResponse` for FastAPI.
- Use `BFastRenderer` for Django Ninja (`NinjaAPI(renderer=BFastRenderer())`).
- Polars and Pandas DataFrames are natively supported by `BFast().encode_packed(df)`.
- In TypeScript, use `bfastFetch<T>(url)` and `bfastQueryOptions` for TanStack Query.
- Refer to the full specification at https://marcelomarkus.github.io/b-fast/llms.txt

4. Custom GPTs / Claude Projects¶

When configuring a Custom GPT or a Claude Project for your team: 1. Download llms.txt. 2. Upload it under Knowledge / Project Files. 3. Your assistant will immediately possess exact knowledge of B-FAST methods and idiomatic patterns.

AI & LLM Integration Guide (llms.txt)¶