CLI & AI Agent Skill
Give your AI coding assistant direct access to your company's data. One install, and Claude Code, Cursor, Codex, and Gemini CLI can query your databases while you work.
Most data questions interrupt your flow. You leave your editor, open a browser, wait for an answer, copy it back. The Dot skill eliminates that loop.
Once installed, Claude Code treats Dot like a sub-agent. It delegates data questions to Dot automatically: "what were our top customers last quarter?" gets answered in seconds, right in your terminal. You never context-switch.
Works with Claude Code, Cursor, OpenAI Codex, and Gemini CLI. The installer detects which agents you have and configures the skill for each one.
Install
Use the URL that matches your Dot region:
US
app.getdot.ai
EU
eu.getdot.ai
macOS / Linux:
Windows (PowerShell):
Then log in:
Opens your browser to authenticate. Token is saved locally.
That's it. Your AI agent can now query your data.
You can also install from the Set Up CLI page in your Dot dashboard (/cli-setup). It generates a command with your auth token embedded so you skip the login step.
Self-hosted Dot:
CI / headless servers:
How it works
The installer puts a native dot binary on your PATH and drops a SKILL.md file into the right location for each AI agent. That skill file teaches your coding assistant when and how to call Dot. It's the bridge between a natural language question and your data.
When an agent invokes the skill, it runs dot as a sub-agent. Dot queries your database and returns a structured answer: text explanation, data preview, chart, CSV, and a link to the full analysis in the browser. The agent reads all of this and uses it in context.
No runtime dependencies. No configuration beyond dot login.
Using with AI agents
Claude Code
Type /dot followed by your question:
Or just ask naturally. Claude Code will delegate to Dot when it recognizes a data question:
"What were our top 10 customers by revenue last quarter?"
"Is the orders table growing?"
"Show me monthly active users for the past year"
Claude Code uses Dot like any other tool in its toolkit. It decides when to call it, reads the results, and incorporates them into its response.
Cursor, Codex, and Gemini CLI
Same behavior. The installer configures each agent it detects. The skill file tells the agent what Dot can do and when to use it. No manual setup.
Standalone usage
You can also use Dot directly from any terminal:
Returns a text explanation, data preview, chart (PNG), CSV data, a link to the full analysis in Dot, and suggested follow-ups.
Follow-up questions
Every response includes a chat ID. Continue the conversation:
View available data
Instant response. Shows your connections, tables, column counts, row counts, and any external assets like Looker dashboards.
Other commands
Caching
Responses are cached on disk. Repeated questions return instantly.
Follow-ups (--chat) and dot catalog are never cached. Cache lives at ~/.cache/dot/, scoped per user.
Security
Tokens stored locally with
600permissionsOne token per user. Generating a new one revokes the old one
Tokens expire after 365 days
All Dot permissions and row-level security enforced
All queries logged for compliance
Troubleshooting
"Not authenticated" Run dot login or check with dot status.
"Connection failed" Check your network. For custom servers: dot login --server https://your-server.com.
Slow responses First query takes 10-30 seconds (full analysis pipeline). Identical queries return instantly from cache after that.
Last updated