dft chat¶
Chat with a terminal AI agent that knows your Dataface project. Sessions are auto-saved to ~/.dft/sessions/ and indexed by working directory.
Arguments¶
| Argument | Description |
|---|---|
PROMPT |
Optional one-shot prompt. Omit to start an interactive session. |
Options¶
| Flag | Description |
|---|---|
--model TEXT |
Model name, optionally provider-prefixed (e.g. openai/gpt-5, anthropic/claude-opus-4). |
--continue, -c |
Resume the most recent session for the current directory. |
--resume TEXT, -r TEXT |
Resume a specific session by id. |
--pick, -p |
Open an interactive picker to choose a session to resume. |
Examples¶
# Fresh interactive session
dft chat
# One-shot question
dft chat "What tables do I have?"
# Resume the last session for this directory
dft chat -c
# Resume a specific session by id
dft chat -r 2026-05-11_14h22m_abc123
# Pick from recent sessions
dft chat -p
Requirements¶
dft chat requires the dataface[chat] extra. Easiest path:
Or install directly:
Set an API key:
What the agent can do¶
The terminal agent uses the same MCP tools that power dft mcp serve — meaning it can:
- Browse your schema (
dft schema,dft schema-search) - Search dashboards (
dft search) - Read packaged YAML docs (
dft docs) - Validate, render, and inspect faces
- Suggest and write face YAML
Related¶
dft init mcp— wire MCP into Cursor / Claude Code / ChatGPTdft mcp serve— run the MCP server directlydft skills— see what skills the agent has loaded