#1 Data Agent that Ships Pipelines Your Team can Trust
Turn hours of dbt and pipeline maintenance, triage, and verification into minutes.
Built by and for people who live in the data stack
You signed up to build pipelines. You spend your day keeping them alive.
Someone renames a column upstream. Models break. The dashboard goes to NA — or worse, a wrong number nobody questions. It's 11pm, and it's your fire.
select- order_total+ order_amountfrom {{ source('shop','orders') }}
“Hey, can you check this number?”
Execs run their own queries in Claude, Slack, ChatGPT — get a confident answer with no provenance — then DM you to confirm it's real. Again. You've become the helpdesk for everyone's AI-generated slop queries.
Slack · direct messageSlack · direct messageSlack · direct messageSlack · direct messageSlack · direct messageSlack · direct messageStop doing the maintenance and the helpdesk. Get back to building.
Point SignalPilot at your stack and it absorbs the two things eating your week.
- ✓scanned 34 sources · resolved 12-node DAG
- ✓fixed stg_orders (order_total → order_amount)
- ✓rebuilt 12 models · dbt tests pass
- ✓verified: row counts · grain · fan-outs
- ✓answered 4 exec queries — with provenance
- ✓28 actions logged — every line tied to a tool call
- ✗ fct_orders — column "order_total" not found
- ✗ mart_dashboard — revenue = NA
- ✗ 6 unanswered DMs: "can you check this number?"
- ✗ no audit trail — can't prove what happened
Triage & maintenance
Reads your whole dbt project, finds what broke, fixes it, and verifies the fix against your real warehouse.
Exec queries
Answers business questions with a receipt — schema-aware, grain-checked — so only the real edge cases reach you.
One governed agent. #1 on Spider 2.0-DBT and ADE Bench.
See the benchmark proofClaude Code is the best coding agent. We make it the best data agent.
SignalPilot is a Claude Code plugin — a governed gateway plus a deep set of data-tuned tools and skills. Same Claude you already trust, now safe and sharp on a production warehouse.
Governance at the wire
DROP/DELETE/ALTER checked by a parser before they run, so the agent can move fast without ever touching prod. Speed and a seatbelt.
Verification built in
40 data-tuned tools and a Verifier that runs 7 deterministic checks against your real warehouse: row counts, grain, fan-outs, nulls.
Persistent context
Your schema, query history, and audit log carry across sessions — so the agent starts every task already knowing your stack.
Same Claude model: ≈14.7% → 65.63% on Spider 2.0-DBT. We didn't replace the agent — we gave it the runtime data work demands.
Describe it. We build, govern, and prove it.
Describe what you need
Tell the agent what to build in plain english.
- Plain-English goal in chat
- Parsed into a plan — no SQL written yet
Agent scans your project
Reads your schema, finds gaps, resolves build order.
- Maps sources, staging, marts, gaps
- Flags date hazards (
current_date,now()) - Resolves build order across the DAG — deterministic, not a guess
Every query is governed
Get rid of unexpected deletes and expensive queries, SignalPilot blocks.
- DDL (
DROP,CREATE,ALTER) and DML (INSERT,UPDATE,DELETE) blocked at the parser - Auto-
LIMIT· budget caps · full audit log
DAG builds itself
Models compile, errors get auto-fixed, tests pass.
dbt parsefirst → materialized in DAG order- Verifier auto-fixes errors · tests loop back
Full audit receipt
Every action logged. Every dollar tracked.
- Duration · turns · queries · blocks · models · columns
- Every line ties to an MCP tool call
One governed engine. Engineers in the IDE, execs in Slack & Notion.
Same governance, same audit log — across Snowflake, Databricks, Postgres, DuckDB and 11 warehouses out of the box.
Same governance. Same audit log. Engineers ship. Execs ask.
It won't touch the rest of your setup.
- ✓It's just a plugin + MCP server — add it, remove it, nothing else changes.
- ✓Works on your existing dbt repo out of the box — no restructuring.
- ✓Or we set it up for you, live, in 30 minutes.
Get back to the work you signed up for.
Book a 30-min setup call and we'll get you running on your own stack — or self-host in 3 commands.
$ git clone https://github.com/SignalPilot-Labs/signalpilot.git$ cd signalpilot && docker compose up -d
$ /plugin marketplace add ./plugin$ /plugin install signalpilot-dbt@signalpilot
$ open http://localhost:3200
Toward an Autonomous Data Stack.
80,000 teams run dbt. As Fivetran–dbt consolidation pushes vendor lock-in, we're building the vendor-neutral alternative — self-healing pipelines that fix migrations before PagerDuty fires, ambient analysts watching your metrics 24/7.