Calabi AI Agent
Professional+
The Calabi AI Agent is a GPT-4o-powered data assistant with access to 117 tools across your entire data stack. Ask questions in plain English — the Agent routes tool calls, synthesizes results, and streams answers back in real time.
How It Works
MCP Tool Coverage
| Server | Prefix | Tools | Capabilities |
|---|---|---|---|
| Calabi Catalogue | om_ | 32 | Catalogue: assets, lineage, tags, tests, glossary |
| CalabiIQ | ss_ | 23 | CalabiIQ: charts, dashboards, metrics, SQL |
| Pipeline MCP | de_ | 49 | Calabi Connect, Calabi Pipelines, dbt, Calabi ML, Calabi Automate, GitLab |
| AWS SDK | aws_ | 13 | EC2, RDS, S3, Cost Explorer |
| Local tools | — | 2 | Chart generation, Excel file generation |
| Total | 119 |
WebSocket API
The Agent exposes a WebSocket endpoint for real-time streaming:
wss://calabi-agent.{domain}/ws/chat
Message Protocol
// Client → Server
{ "message": "Show failed Calabi Connect syncs today", "history": [] }
// Server → Client (streamed chunks)
{ "type": "text", "content": "Checking Calabi Connect..." }
{ "type": "tool_call", "name": "de_list_jobs", "args": { "status": "failed" } }
{ "type": "tool_result", "name": "de_list_jobs", "content": "[...]" }
{ "type": "text", "content": "Found 2 failed syncs..." }
{ "type": "done" }
// Download response
{ "type": "download", "content": "<base64>", "filename": "syncs.csv", "mime": "text/csv", "rows": 47 }
Example Queries
# Data discovery
"What tables contain candidate data?"
"Find all columns tagged as PII"
"Who owns the psychometric_scores table?"
# Pipeline monitoring
"Is the data unification DAG running?"
"Show me all failed Calabi Connect syncs in the last 24 hours"
"What caused the dbt run to fail this morning?"
# Analytics
"Run a SQL query: count unique exams by day this week"
"Create a bar chart of pass rates by country"
"Export exam results for Q1 2026 as Excel"
# Data engineering
"Trigger a full refresh of the Salesforce connection"
"Scaffold a dbt model for the candidates table"
"Create a data quality test for unique exam IDs"
# ML
"What's the best run in the psychometric-scoring experiment?"
"Register run abc123 as the production model"