Data Context Layer
A context layer for your company’s data
Shanc builds a structured context layer from your warehouse schema, git history and query logs — your metrics, entities and the rules that connect them — and serves it prebuilt to any agent over MCP.
- Built from three signal types: schema, git history, query logs
- Structured definitions: metrics, entities, relationships, business rules
- Served over MCP to any agent your team uses
How it works
Built from your systems
Your business logic, read from where it lives
Shanc reads warehouse schema, dbt models and git history, and the queries your analysts actually run. The layer is assembled from how your company already works with its data — connecting a source is the whole setup.
Structured definitions
What’s inside the layer
Metrics with their exact calculation logic. Entities and how they join. Relationships between tables, and the business rules that decide which revenue is “revenue.” Ambiguities get resolved into one definition, so every agent works from the same answer.
Served over MCP
Prebuilt context in every agent
Agents receive definitions ready to use — in Claude, Cursor, Copilot or your own stack. Answers start from your company’s real definitions rather than a fresh guess at your schema on every question.
95% accuracy on a real 118-table warehouse — against 73% for agents working from raw schema.
View the benchmarkContext decides whether agents make it to production
On Spider 2.0, built from real enterprise warehouses, the best agent frameworks solve 21.3% of tasks — the same class of models that scores 91.2% on academic SQL benchmarks. Raw schema access gives agents tables without meaning. (Spider 2.0, Yale/Stanford et al.)
76% of data experts say 20–49% of their team’s time goes to preparing ad-hoc reports for business teams. Today, the context layer is people. (Sigma Computing survey)
Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls. The survivors will be the ones whose agents can be trusted with real questions. (Gartner, June 2025)
A context layer turns your warehouse from something agents query into something they understand.