SEC AI Tool Router & Query Planner

One structured SEC goal in; an explicit, schema-aware, cost-bounded tool decision out. The Actor creates deterministic secAiRunPlan records for one to twenty controlled SEC Complete Data Suite intents.

An AI client gets the exact Actor input, expected record type, availability, fallback disclosure, and minimum/maximum published plan charge. It can approve or reject each visible step instead of guessing from titles or loading 36 product pages into its decision prompt. executionPolicy remains clientExecutes; when the caller is an AI agent, that agent is the orchestrator.

The planner reads only its versioned build-time catalog. It does not execute a plan, start another Actor, access SEC, accept free text, or use an LLM, embeddings, or vector search. It therefore spends zero model tokens internally. A client using the compact Suite catalog avoids README-selection tokens by construction; further token savings from structured or compact records depend on model, tokenizer, source, and evidence settings and are not given a fixed percentage.

The Suite-level agent-frontdoor-input.schema.json and deterministic compile_frontdoor_request.py provide a strict reference projection from the six outcome IDs to this Actor's executable input schema. They perform no network or Actor call.

Use canonical CIK, ticker, accession, series, class, or other accepted exact identifiers with a closed period, record ceiling, and USD budget. A capability that is unavailable without a public fallback produces CAPABILITY_NOT_AVAILABLE, never an invented plan.

This Actor is not investment, legal, or compliance advice.