Dwata
Connect your email inbox, download emails, and use LLM agents to extract financial transaction data — all running locally on your machine.
All data stays on your machine. dwata works with Ollama and local models (tested on Mac Mini M4 16GB), so your emails never leave your computer.
[!WARNING] dwata is very early software and is being developed actively, I am sorry if the extracted data has bugs.
What dwata does today
Email inbox
Connect your Gmail or IMAP account. dwata downloads your emails and stores them locally in SQLite.
Financial template detection
Select financial emails and run an LLM agent to generate extraction templates. The agent reads sample emails and produces reusable patterns — you only need AI once per email sender.
Financial templates
Browse and manage the generated templates. Each template captures how to extract financial data from a specific sender.
Extracted transactions
Once templates are in place, dwata extracts financial transactions from matching emails automatically.
[!WARNING] There are quite a few issues with the extraction logic. I am working on it actively.
LLM settings
Use Ollama with a local model (Ministral 3: 3b), OpenAI (GPT-4o Nano), or Google Gemini (Gemini 2.5 Flash Preview). Switch models in settings.
Privacy
- All email data is stored locally in SQLite — never sent to a cloud service
- With Ollama, template detection runs entirely on your machine with no external API calls
- OS keychain stores credentials (Gmail password / OAuth tokens) — dwata uses a single keychain entry so you get one prompt on first launch
- On macOS, select Always Allow when the keychain prompt appears so you're not asked again
Getting started
Download the latest release for your platform from GitHub Releases.
Launch the Dwata desktop app (Tauri). It starts the dwata-api sidecar automatically and loads the GUI inside the app window.
Connecting Gmail
dwata supports Gmail via OAuth. Set your Google OAuth client_id and client_secret in the config file:
- macOS:
~/Library/Application Support/dwata/project.toml - Linux:
~/.config/dwata/project.toml - Windows:
%APPDATA%\dwata\project.toml
You can use your own Google OAuth app (bring-your-own credentials).
Using Ollama
Install Ollama and pull model Ministral 3:3b:
ollama pull ministral-3:3b
Then set the model in dwata's settings page.
Tech stack
- Backend: Rust + Actix-web
- Database: SQLite (local)
- Frontend: SolidJS
- UI Components: DaisyUI
- LLM: Ollama (local), OpenAI, or Google Gemini
Support
- Bugs & issues: GitHub Issues
- Discussions: GitHub Discussions
- Developer guide: DEVELOP.md
License
GPL v3 — see LICENSE.
From the founder
I am Sumit, and I live in a small eastern Himalayan village in India. I mentor/co-mentor hundreds of folks each month about how to use coding agents. I run a digital nomad space in our village. Come, say Hi!