Dwata

Dwata is a local-first Rust backend for connecting email accounts, storing mail in SQLite, and using LLM agents to extract financial data and knowledge-graph entities. Email data stays on the machine running dwata-api unless you choose an external AI provider.

Warning

Dwata is an active restart. The native cross-platform UI is planned but is not shipped yet. This repository currently provides the backend API and Rust agent components; it does not provide a downloadable desktop application.

What is available now

• dwata-api: an Actix-web JSON-over-HTTP API with local SQLite storage, email/OAuth integration, search, download jobs, and extraction endpoints.
• dwata-agents: Rust agents for financial and knowledge-graph extraction.
• shared-types: Rust domain and API types shared by the backend and agents.

The API is intended to be the server boundary for the restart. A future native Rust UI will compose with the same application capabilities; the former web and Tauri desktop applications are no longer part of the project.

Run the backend

Install the Rust toolchain, then start the API from the repository root:

cargo run -p dwata-api

By default it listens on 127.0.0.1:8080. Verify that the service and SQLite connection are available:

curl http://127.0.0.1:8080/api/health

The first start creates a configuration file in the OS configuration directory when neither config.toml nor ../config.toml exists:

• macOS: ~/Library/Application Support/dwata/config.toml
• Linux: ~/.config/dwata/config.toml
• Windows: %APPDATA%\dwata\config.toml

Copy and adapt config.example.toml if you need to set the listen address, OAuth client credentials, AI-provider keys, or search index path. See the developer guide for configuration, database, and agent-development details.

Gmail and AI providers

Gmail OAuth uses a Google OAuth client configured under [google_oauth] in config.toml. The redirect URI is derived from [server] as http://<host>:<port>/api/oauth/google/callback; register that exact URI with your OAuth application.

For local inference, install Ollama and pull the default model:

ollama pull qwen3.5:2b

OpenAI and Gemini keys can instead be configured in [ai_provider_api_keys]. Using an external provider sends the applicable extraction prompts and content to that provider.

Privacy and storage

• SQLite data is local. Its default path is OS-specific; see DEVELOP.md.
• Credentials are stored as SQLite metadata plus secrets in the OS keychain.
• Ollama inference is local; third-party AI providers are not.

Development

Backend development needs Rust. SQLite's CLI is optional for inspecting the database. Node.js, TypeScript, Tauri, and a web GUI are not required or included in the current workspace.

Useful commands:

cargo fmt --check
cargo test --workspace
cargo build --workspace

Support

• Issues
• Discussions
• Developer guide

License

GPL v3 — see LICENSE.