fireSeqSearch

Local semantic search and RAG over your Logseq or Obsidian notes, surfaced in your search engine.

When you google, fireSeqSearch appends hits from your personal notebook to the search results, and optionally asks an LLM to answer your question grounded in those notes. All local.

Works the same on Bing, DuckDuckGo, Searx, and Metager — "google" is just shorthand.

Want the pre-LLM version? Use backend release 0.9 with the latest addon. Details in docs/README-pre-llm.md.

More examples at docs/examples.md.

What you get

• Semantic search — finds your notes by meaning, not keyword overlap.
• One-line LLM summary per page, so you can scan results at a glance.
• /ask Q&A — ask a question, get a cited answer drawn from your notes.

Installation

Install bottom-up: LLM backend → local server → browser extension. The extension is useless until the server is running, and the server is useless until the LLM backend answers.

1. Local LLM backend

The server talks to an OpenAI-compatible HTTP backend for embeddings and chat. The embedding model is bge-m3 (1024-dim, multilingual) — chosen and pinned for retrieval quality. Any reasonable instruct-tuned chat model works.

Embedding is zero-config. On first run the server auto-downloads a pinned, self-contained bge-m3 llamafile (~723 MB) into ~/.cache/fire_seq_search (verified by SHA-256) and launches it for you. There's nothing to install for embeddings — the only model you choose is the chat model.

Drop the chat GGUF in ~/llm/ — that's where the server looks by default:

• ~/llm/Qwen3.5-9B-UD-Q4_K_XL.gguf (chat)

Override the chat model with --chat-model if you keep it elsewhere. To use your own embedding model instead of the auto-downloaded one, pass --embed-model /path/to/model (GGUF or llamafile).

By default the server spawns its own llama-server; see build_llama_server.sh and Containerfile for the Vulkan build. To use an existing server (Ollama, remote llama), pass --embed-endpoint / --chat-endpoint.

2. Local server

Install Rust: https://doc.rust-lang.org/cargo/getting-started/installation.html

Min Rust version: see .github/workflows/rust.yml.

git clone https://github.com/Endle/fireSeqSearch
cd fireSeqSearch/fire_seq_search_server
cargo build --release

Logseq

./target/release/fire_seq_search_server --notebook_path /home/you/logseq_notebook

Or use debug_server.sh as a template.

Obsidian

./target/release/fire_seq_search_server --notebook_path /home/you/vault --obsidian-md

Or use debug_obsidian.sh as a template.

The server hosts endpoints on http://127.0.0.1:3030. The extension talks to it from your browser.

3. Browser extension

Firefox only: https://addons.mozilla.org/en-US/firefox/addon/fireseqsearch/

Example

AstroWiki-RAG-2026-05-24.webm

Notebook provider: AYelland/AstroWiki_2.0

License

MIT (both server and addon). Third-party libraries may have their own licenses; see source.

LOGO: https://www.flaticon.com/free-icon/web-browser_7328762 — Flaticon license. UI icons by manshagraphics — Flaticon.

Similar projects

• karlicoss/promnesia — broader scope; fireSeqSearch only appends notebook hits to search results.
• Logseq Copilot

How it works

  notes on disk                    local LLM backend
  (Logseq / Obsidian)              (llama-server / Ollama)
        │                                  │
        ▼                                  │
  chunker  ────────► embeddings ◄──────────┤
        │                                  │
        ▼                                  │
  SQLite store ──► in-memory cosine        │
        │                                  │
        ▼                                  │
   /query  /ask  ◄───────── chat ──────────┘
        │
        ▼
  browser extension appends to search results

• Index: ~10K chunks fit in a flat in-memory Vec<[f32; 1024]>, brute-force cosine. No vector DB, no ANN.
• Storage: SQLite holds notes and chunks; the index is rebuilt in memory from SQLite on startup.
• Refresh: mtime + Blake3 content hash detect changes. 10-minute background rescan; manual POST /reindex trigger.
• LLM serving: OpenAI-compatible HTTP (embed + chat). By default the server spawns its own llama-server; you can also point at a pre-running server (Ollama, remote llama) via --embed-endpoint / --chat-endpoint.

See CLAUDE.md for the locked technical decisions and rationale.

Star history

Provided by https://star-history.com