Agent integration

StrataBI's dashboard contract is documented machine-first so autonomous agents can generate valid dashboards without scraping these pages. This page is the entry point for that workflow.

Machine-readable resources

The dashboard schema is JSON Schema 2020-12 and carries embedded instructions (x-llm-instructions) describing how a generator should treat it.

Per-module instructions

A deployed module may ship an llm_context.txt — a short, module-scoped manual at modules/<module_id>/llm_context.txt describing the module's functions, their inputs and outputs, and an example tile. Before wiring a lambda tile to a module, fetch that module's llm_context.txt and follow it. See Build a module.

Generating a dashboard

The contract is intentionally narrow: a dashboard declares intent only. Infrastructure resolution, module lookup, source-value lookup, dependency ordering, trigger propagation, result loading, and execution orchestration are all owned by StrataBI core — an agent should not try to produce them.

A reliable loop:

1. Fetch dashboard.schema.json.
2. Generate a dashboard object that conforms to it. Use only documented

block types, exec types, and load modes.

1. Validate the output against the schema. Reject and retry on any error.
2. Save the validated JSON.

fetch schema ─> generate JSON ─> validate against schema ─> save
                      ▲                     │
                      └──── retry on error ─┘

Rules of thumb for generators

• Output valid JSON only — no comments, no trailing commas.
• Reference inputs by input_id; bind them into SQL with query.params.
• For Athena tiles, always include query; for lambda tiles, include

module_id and lambda_index.

• Don't invent fields. additionalProperties is closed in most of the schema, so

unknown keys fail validation.