SignalPilot

What is SignalPilot?

SignalPilot is a governed AI agent for dbt-native data teams. It ships grain-safe dbt models, self-heals broken pipelines, and answers metric questions in Slack — all through a single 7-layer governance engine that reads your schema.yml, macros, and sources before writing a single line of SQL.

How does SignalPilot's 7-layer governance gateway work?

Every query flows through seven fail-closed layers: AST validation, automatic LIMIT injection, cache check, budget enforcement, read-only execution, PII redaction, and immutable audit logging. DDL statements (ALTER, DROP, etc.) are blocked at the outer shell, so agents can never mutate or destroy production data.

What is SignalPilot's score on Spider 2.0-DBT?

SignalPilot scored 51.56 on the Spider 2.0-DBT benchmark — #1 on the leaderboard, +7.45 over second place (Databao/JetBrains at 44.11), and the first entry ever to cross 50% in the benchmark's 11-month history. The result comes from harness architecture — the 7-Check Verification Protocol run by a dedicated Verifier subagent, and the dbt-aware static scanner — not just the underlying model.

Which data stacks does SignalPilot support?

SignalPilot is vendor-neutral across warehouses — Snowflake, BigQuery, Databricks, and Redshift — and dbt-native. It reads dbt Core and dbt Cloud projects, including schema.yml contracts, macros, sources, and packages.

Is my data ever used for training?

No. SignalPilot processes queries through the governance gateway with PII redaction applied before any data leaves your environment. Customer data is never used to train foundation models, and enterprise deployments support fully isolated, tenant-specific inference.

How do I install SignalPilot?

Install via the Claude Code plugin registry: claude plugin install github:SignalPilot-Labs/signalpilot. The plugin connects to your dbt project and warehouse via read-only credentials.