Understand any codebase. Ask what matters.
Sonar creates local-first, source-grounded codebase briefings for people who need project context without a subscription to a frontier coding model. It helps product managers, founders, designers, support leads, operators, executives, and technical teammates understand what a repository does, where important workflows live, and what to ask next.
Sonar is optimized for high-level orientation. For fine-grained debugging, refactoring, line-by-line code explanation, or implementation work, use an engineer or a frontier coding tool with full repository context.
- Generates source-grounded briefings for an indexed repository.
- Supports session-aware follow-up questions for orientation, risks, workflows, systems, and source navigation.
- Builds a deterministic repository inventory, asks the model to inspect selected source files, and stores a source-backed memory graph before writing the initial briefing.
- Uses exact lookup, grep-like lexical search, graph expansion, workflow planning, and briefing-specific ranking for follow-up questions and source lookup.
- Returns source lists and citation verification diagnostics with generated answers.
- Persists projects, source files, generated memory graphs, briefing sessions, and generated briefings in SQLite.
- Runs as a Tauri desktop app backed by a local workspace engine and embedded SQLite project store.
- Supports local llama.cpp/OpenAI-compatible generation or custom OpenAI-compatible cloud/local model endpoints.
- Lets users copy or export generated briefings as Markdown.
Sonar is intended to produce strong source-grounded briefing drafts, especially with local models. It is not a replacement for human-reviewed technical, compliance, security, or architecture documentation.
Sonar currently indexes TypeScript/TSX, JavaScript/JSX, Python, Rust, Go, Java, C#, Ruby, C++, PHP, Kotlin, Swift, Markdown/MDX, plus selected JSON and Prisma schema files as text evidence. The repository survey also scans common source extensions for high-level signals such as entry points, file IO, network boundaries, config, logging, and state. Repositories that mainly use unsupported source languages can still be imported, but briefings may be less complete because full parser coverage is not available for those files.
See Language Support and Limits for the current parser coverage and known limitations.
Sonar is currently distributed as a source-built desktop app. The expected local setup is:
- Node.js 22.x, 23.x, 24.x, or 25.x. Do not use Node 26+ for source builds.
- Rust toolchain and platform dependencies for Tauri
- Git, if you want Sonar to clone GitHub repositories
- A local OpenAI-compatible model server, or an OpenAI-compatible API endpoint you explicitly configure
Clone the repository and install dependencies:
git clone https://github.com/gautamdhameja/sonar.git
cd sonar
nvm use 24
npm installIf you previously installed dependencies with a different Node major version, run npm install again after switching
Node versions. Native dependencies such as SQLite are compiled for the active Node runtime.
Start a model server separately. For local mode, Sonar expects an OpenAI-compatible endpoint such as llama.cpp
llama-server on http://127.0.0.1:8080/v1, and the endpoint must respond to /models.
When the server also exposes llama.cpp-style /props with n_ctx, Sonar sizes its source-context budget dynamically:
small context windows get compact prompts, while large-context models receive more repository evidence up to Sonar's
interactive local-generation cap.
Build the production desktop app:
npm run desktop:buildOn macOS, open the built app:
open src-tauri/target/release/bundle/macos/Sonar.appOn first launch, Sonar starts its local workspace engine, asks whether you want a local model endpoint or an
OpenAI-compatible API endpoint, and stores project data in embedded SQLite under ~/.sonar. It does not require
Meilisearch or any other external database/search service.
This source-built alpha still uses the checked-out repository to run the local workspace engine, so keep the checkout and
node_modules in place after building. npm run desktop:dev is for contributors who need the hot-reload development
shell, not for normal local use.
For detailed setup steps, see Setup from Source.
Sonar indexes only the GitHub clone or local folder you explicitly select in the desktop UI. Project state, generated briefings, memory graphs, and runtime settings are stored locally under ~/.sonar.
Desktop model settings are stored locally at:
~/.sonar/desktop-config.json
Do not commit this file. It may contain API keys.
If you use API endpoint mode, source excerpts needed for generation are sent to the configured model provider. If you use a local model endpoint, repository analysis stays on your machine.
Detailed documentation lives in docs/.