Solutions · Self-improving data contextData context that keeps itself correct.
An agent authors the semantic context for every data model from your dbt project, dashboards, and queries, proves it with evals, and folds every correction back in. You approve; nothing changes silently.
How it runs
Parse, author, prove, learn.
Four steps the agent repeats on every change to your data models, with your approval as the gate.
- 01
Parse
The agent reads your schemas, dbt models and tests, trusted dashboards, and the queries behind them, and traces how each metric is actually calculated. Missing documentation, conflicting definitions, and likely data issues surface here.
- 02
Author
Semantic context per data model: meaning, calculation, sources, exceptions. Written as Markdown and Apache Ossie definitions next to the models, with required dbt changes opened as pull requests.
- 03
Prove
Graded questions generated for each definition, anchored in a gold value and the approved definition, failure cases included. The suite reruns on every change and gates the pull request.
- 04
Learn
Thumbs-downs, new questions, and drift become traced proposals. The agent updates the context and the evals; you approve, and the correction becomes part of the system.
Every change carries its trace
Each proposal is a pull request that links the feedback, the thread, the queries, and the eval result that motivated it.
- Only the evals the change touches rerun, and the blast radius names the models and dashboards affected.
- You approve; the merge unlocks only when the numbers hold.
- Context lives in Markdown and Apache Ossie in your repo, readable by your BI tools.
feat: refund reason on int_orders_with_refunds #212
signalpilotproposed · feat/refund-reason → main
- add refund_reason to int_orders_with_refunds
- fix: keep 1:1 grain on order_id · signalpilot
- ✓Dana R. approved these changes
Keeping it honest
Coverage is the health metric.
Two numbers on one dashboard: how many of your models have graded questions, and how often those questions passed in the last 30 days. Together they tell you whether the context can be trusted and what to cover next. The more models with graded questions, the better the answers.
- Definitions in Markdown and Apache Ossie, next to the models.
- A coverage dashboard and the eval suite, in your repo.
- Blast radius on every pull request.
Bring your dbt project. Leave with context that maintains itself.
Open source · Apache 2.0 · GitHub · Slack