Skip to main content
Docker runs the web app, Postgres, a background worker and synthetic demo records. Install Docker with Compose; on Windows, use a WSL 2 terminal. You do not need to install Python or Node.js, and a GPU is optional. Mirobody is available under Apache 2.0 on GitHub.

​
First answer in three steps

1

Start the stack

The script uses OpenSSL to generate secrets in .env, downloads the application image, starts the demo environment and prints a setup link.
2

Choose a model

Open the printed link and paste one model key: OpenRouter, OpenAI, Gemini, Anthropic, DeepSeek, DashScope, or any OpenAI-compatible gateway. The key is kept only after one real request through it works, and is stored encrypted; Settings › Model changes it later, without a restart. To keep every model on this machine instead, see Where the models run.
3

Sign in and ask

Open and sign in as you@mirobody.ai with code 111111 on the Email code tab. Ask “How has my cholesterol changed?”: the answer charts the trend and names the file behind each number. Then upload a file from demo/upload/ on the Data page, or switch to the record mom@mirobody.ai shares with you.
The walkthrough follows the same record through five scenes. For real records, apply the settings in Deploy on a Server before the first start; after a demo, follow the existing-demo path to start with clean data volumes.
From the engine checkout:
The mirobody service should be healthy, and the health response includes version and agent. mirobody doctor names the entry each model surface selected; --probe sends each a real request. A key added to .env later takes docker compose up -d; docker compose restart does not re-read .env. Troubleshooting lists the common failures.
Download and extract the source archive instead:

​
Where the models run

To run the models in Docker beside the stack, start it with a Compose profile instead of the plain ./deploy.sh:
These profiles configure the connection automatically. The first request waits for the model download. On a Mac, run llama-server on the host to use the GPU. See Run Locally for platform setup and Choosing a Model for the available options. A cloud API key in .env takes priority; use the overrides in Configuration to select local models explicitly.

​
What leaves the machine

Records stay in your Postgres database. Mirobody does not send usage telemetry. External data transfers depend on the services you enable: Encryption at rest covers chat, uploaded files, medication text and your profile, not yet readings or genotypes. The app listens on 127.0.0.1 only; before it reaches a network you do not control, read , which lists every address the server can call.

​
Export and backup

Readings export as CSV or JSON, the visible record streams as NDJSON, genotypes export as VCF or FHIR, and shell/backup.sh makes the complete copy of the database and uploaded files. Export Your Data has the routes and their limits.

​
Next steps

Connect your agent

Claude Code, Codex, Cursor or Gemini CLI over MCP, with a personal link from Settings.

Use your own data

Reports, photos and spreadsheets, then devices and Apple Health exports.

Configuration

Model keys, local models, configuration files and deployment settings.

Deploy on a Server

Production settings, secrets, HTTPS, upgrades and backups.
To resolve names and units in your own code without the stack, see the library. A self-hosted engine serves the web client at /, its own HTTP API at /api/* and MCP at /mcp.

​
Contributing

The most useful contribution is a term the resolver gets wrong. Run mirobody resolve "<term>"; if the answer is wrong or empty, report it, or add a row to together with a case in . The full workflow is in ; security reports go through a private advisory, as describes.