⚡ SignalPilot Data Agent

SignalPilot offers governed AI agents for data stacks, providing secure, production-ready access to databases and dbt projects. Targeted at Agentic Data and Platform Teams, it delivers enterprise guardrails and optimization, preventing common data issues like hallucinations, dropped rows, or unbounded queries, and achieving state-of-the-art benchmark performance.

How It Works

SignalPilot employs a five-stage workflow: Plan (natural language goal), Scan (dbt project and warehouse inspection for hazards), Govern (enforcing policies like LIMIT injection, DDL blocking, budget caps with detailed auditing), Build (topological DAG materialization with a verifier agent for fixes), and Report (generating a structured audit receipt). Core infrastructure includes an AI assistant Plugin, an MCP server for communication, and an Observability platform for monitoring and management, prioritizing safety, determinism, and auditability in AI-driven data operations.

Quick Start & Requirements

Self-hosting SignalPilot is initiated via Docker Compose: git clone https://github.com/SignalPilot-Labs/signalpilot.git, cd signalpilot, docker compose up -d. The system requires Docker and a compatible AI agent (Claude Code is recommended for full skill integration). Supported databases include DuckDB, PostgreSQL, SQLite, Snowflake, and BigQuery. A free cloud trial is available at https://app.signalpilot.ai/sign-in, with comprehensive documentation at https://SignalPilot-Labs.github.io/SignalPilot/.

Highlighted Details

• Achieved #1 ranking on the Spider 2.0-DBT benchmark (51.56% score).
• Features 32 MCP tools covering query execution, schema discovery, dbt intelligence, and model verification.
• Implements robust governance: automatic LIMIT injection, DDL/DML blocking, dangerous function denylists, per-session budget caps, and comprehensive query auditing.
• Includes a verifier agent that analyzes dbt build outputs, detects date hazards, and proposes fixes for model issues.

Maintenance & Community

Community engagement occurs via GitHub Discussions for questions and ideas, and open issues for bug reports and feature requests. The project encourages community contributions and stars. Further details can be found at https://github.com/SignalPilot-Labs/signalpilot/discussions.

Licensing & Compatibility

SignalPilot is licensed under the Apache 2.0 license, which generally permits commercial use and integration into closed-source projects without significant copyleft restrictions.

Limitations & Caveats

While the core MCP tools are broadly compatible with various AI clients, the advanced "skills" and orchestration logic are specifically designed for Claude Code. Using SignalPilot with other MCP clients is considered experimental and may lack the full functionality of the Claude Code integration.