Build context that never goes stale.
Context only works if it’s available everywhere, from everywhere, and kept up to date. Build a context factory that updates as it’s used, no matter where the knowledge lives.
The context production loop
Build a context layer that improves
every time someone uses it
01 / 04
Assemble
Assemble a context layer from what you already have.
- Ingest dbt YAML, docs, past queries, and the answers your team repeats
- Generate one versioned context layer for the whole catalog
- Own portable YAML that moves across systems
Context layer
v12The context production loop
Build a context layer that improves
every time someone uses it
01 / 04
Assemble
Assemble a context layer from what you already have.
- Ingest dbt YAML, docs, past queries, and the answers your team repeats
- Generate one versioned context layer for the whole catalog
- Own portable YAML that moves across systems
Context layer
v1202 / 04
Monitor
Monitor every interaction with your context layer in one place.
- See context interactions across chatbots, MCPs and APIs
- Collect key usage data in one place
- Identify the questions agents get wrong, and why
Monitoring
Every interaction with your context layer
1,284
612
448
224
03 / 04
Fix
Fix a wrong answer at its source.
- Diagnose which definition produced it
- Draft a one-line change in the layer’s own YAML
- Send the change for review before anything ships
Fix · #008391
Why did this check fail?
Thought 3s
Read eval set #008391
Traced investment_grade
The boundary excluded BBB− itself. Widened it and left the grain alone.
Metrics investment_grade
10metrics:
11investment_grade:
12 expression:
13 credit_rating > ‘BBB−’
13 credit_rating >= ‘BBB−’
14 grain: issuer
04 / 04
Govern
Govern every change with a version you can roll back.
- Review each change as a diff before it lands
- Version the layer on every approved change
- Roll back to any earlier version when you need to
− credit_rating > ‘BBB−’
+ credit_rating >= ‘BBB−’
Context layer
New · v13investment_grade = credit_rating >= ‘BBB−’
The failed question joins the eval set
Built to fit in
Interoperable
by design
Your context layer plugs into the store you already run, travels as open standards, and lands inside your existing workflows.
Any context store
Works with any
context store
Point it at your data warehouse, metric store, internal PDFs, or agent platform. The layer reads from your context where it lives.
Open standards
Open standards,
portable anywhere
The layer is portable YAML: models, metrics, and relationships you can read, diff, and move. It is yours to take wherever your stack goes.
Your workflows
Deployed into
your workflows
Our forward-deployed engineers build it alongside your team and ship it into the tools you already use, so there is no big migration.
How we work
Benchmark it. Spec it. Build it.
We start with a short, scoped benchmark of this practice: where you stand, against whom, and what to build first. You own everything we produce, with us or without us.
Benchmark
- Run real agent queries against your data
- Score how much context an agent has to be handed
- Trace where context lives, and who keeps it current
Spec
- Design the target context layer
- Inventory every context source worth assembling
- Set what runs automatically, and what asks first
Build
- Assemble the context layer with agents
- Stress-test it with real customer questions
- Ship it versioned, with review and rollback
Manage
- Background agents propose improvements from usage
- You review the diffs and decide what ships
- Questions agents get wrong shrink, release over release
Customers
The infrastructure powering the data economy
The context benchmark
See your data the way an agent sees it.
We run real agent queries against your products, score how much context an agent has to be handed, and trace who keeps yours current today. You keep the readout, the gap list, and the build plan.
Request the benchmark