AI assistant

The assistant is optional and off until a project owner adds an API key. It runs on your own key, from the provider you pick, and nothing goes to a model you didn’t choose.

Turn it on

1. Get an API key from a provider: Anthropic, OpenAI, Google Gemini, OpenRouter, Mistral, xAI, DeepSeek, Groq, or any OpenAI-compatible service.
2. As a project owner, open Project settings›Assistant.
3. Pick the provider (it’s guessed from the key), paste the API key, and pick a model.
4. Optional: give the assistant a name. It answers to it.

The key is checked, then stored encrypted. Only its last characters are ever shown. Usage is billed to that key.

Use it

• Open Assistant in the icon bar and ask about the project. It reads objects, docs, graphs, tables, tasks, milestones and assets.
• Type @ to point it at a file or a teammate.
• It can propose changes (new tasks, doc edits, object fields), which you apply from the chat.
• In a doc, select text and Ask about it, or have it suggest a rewrite.

Answers are worked out on the server, so they keep going when you leave the page.

Run a model on your own machine (self-hosted)

Use Ollama, LM Studio or any OpenAI-compatible server on your network:

1. On the Qwest server, set ALLOW_PRIVATE_AI_ENDPOINTS=true in .env and run docker compose up -d.
2. In Project settings›Assistant, pick Other (OpenAI-compatible) and enter its endpoint, like http://192.168.1.50:11434/v1.

From inside Docker, localhost is Qwest’s own container: use the machine’s network address instead.