Your notes. Your models. Your machine.

AI chat over your Obsidian vault — local only.

Ask questions in natural language. Prometheus searches your notes by meaning, answers only from what it finds, and shows the sources — with no cloud LLM or embedding APIs.

Community Plugin + local sidecar Runtime + Ollama. Desktop only.

Get started How it works
Local First Offline First Apache-2.0 Ollama

Capabilities

One plugin — full local RAG

UI in Obsidian, heavy work in the managed Runtime, generation through your Ollama model. Everything stays on your machine.

Grounded chat

Answers only from retrieved vault chunks. If context is thin, Prometheus says so instead of inventing facts.

Semantic search

Find notes by meaning, not only exact keywords. Retrieval works without generating a full chat reply.

Cited sources

Every answer links back to notes and headings. Open the source in Obsidian and verify it yourself.

Automatic indexing

Markdown is chunked and embedded locally. Later edits update the index incrementally.

Note context

Add specific notes to the prompt with the / picker when you need tighter context.

Search scopes

UI scopes: Auto, whole Vault, current note, or current folder — narrow retrieval to the task.

Streaming & cancel

Answers stream in. Cancel generation, restart the runtime, or reindex the vault when needed.

Managed Runtime

The plugin downloads and runs the sidecar outside the vault, verifies checksum and version — or point it at your own binary.

Any Ollama model

Pick any local model you already have installed. Prometheus is not locked to a single LLM.

Interface

Obsidian chat with cited sources

Ask about Git, architecture, or your own notes — the answer is grounded in the vault, not the model’s general knowledge.

Prometheus chat answering a question from vault notes with cited sources

How it works

From question to cited answer

Three local processes: a thin Obsidian plugin, Prometheus Runtime, and Ollama.

1

Semantic search

Runtime embeds the query and retrieves relevant chunks from sqlite-vec (optional hybrid + rerank).

2

Grounded prompt

Only retrieved fragments enter the context. Grounding policy forbids filling gaps from model trivia.

3

Local generation

Ollama generates on localhost. The plugin streams the text and shows sources you can open.

Obsidian Community Plugin  (TypeScript UI)
            │  127.0.0.1
            ▼
Prometheus Runtime  chunking · FastEmbed · sqlite-vec · retrieval
            │  localhost
            ▼
         Ollama → local LLM
            │
            ▼
     Answer + note sources

Full architecture walkthrough

Privacy

After setup, Wi‑Fi can stay off

Base mode uses no cloud LLMs, cloud embeddings, or third-party note uploads.

Stays on your device Not used in base mode
Vault Markdown Cloud LLM APIs
Embeddings & vector index Cloud embedding APIs
Chat prompts & answers Third-party note upload
Ollama model weights Hosted “Prometheus cloud”

Documentation

Features and setup in depth

From quick install to RAG pipeline details and FAQ.

Setup

Ollama, plugin, managed Runtime, Ready status, and your first question.

How it works

Process architecture, indexing, retrieval, and grounded generation.

Features

Chat, Search, scopes, Settings, Status, diagnostics, and 0.1 boundaries.

FAQ

Offline use, mobile, cloud, custom binary, and “not in my notes”.

Prometheus is yours.

Open source, Apache-2.0. No sales team, no demo request.

Install GitHub repository