omnifeed

omnifeed docs

omnifeed lets an AI agent search the web and read the results, on self-hosted SearXNG and crawl4ai. Its Reddit engine returns full comment trees as TOON.

Install and quick start are in the README.

Guides

• Configuration: every OMNIFEED_ variable, content-size controls, Reddit size limits and Prometheus metrics.
• omnifeed on Apple container (macOS): run the stack on Apple’s native container runtime instead of Docker.
• Rate limits: how many searches and fetches the upstream engines and sites tolerate, measured per minute, hour, day and month, with the block log and the queries to check your own deployment.

Reference

• Error codes: the stable error codes every front-end reports, how failure reasons map onto them, and the MCP isError result shape.
• API reference: Go package docs generated from source.
• Ideas and parked work: features considered and not built, or built and removed, with the measurements, what would justify revisiting each, and the git ref that holds the code.

Decisions

• ADR 0001: Reading Reddit after the logged-out JSON shutdown: why the Reddit engine uses a client_id → session → browser method chain, with every option measured or rejected, and how to refresh the session cookie.
• ADR 0002: Waiting out Cloudflare challenges with a CSS selector: why the challenge wait is a selector since crawl4ai 0.9.5, the options rejected, and the order to roll it out in.
• ADR 0003: go fix modernizers are enforced in CI: make fix-check, and what a new Go release means for it.
• ADR 0004: Releases are signed keylessly with cosign: what is signed, with which identity, and how to verify it.

This site is open source. Improve this page.