Concepts

Decisions

A decision is the central object — a title, a body of context, and (once analyzed) a set of framework stages. Every tool call that touches a decision needs its decision_id, returned when you create one.

Frameworks

Pass a framework_hint when starting a decision, or omit it — DACI is the default. Call list_frameworks / GET /v1/frameworks for descriptions if you want a planner to choose. Not sure which fits? assess_decision reads a decision's saved context and recommends a complexity level, framework, and title — review it, then commit with apply_decision_setup (as-is or overridden).

Each framework analysis produces an ordered set of stages (e.g. DACI: Driver, Approver, Contributors, Informed). get_recommendation returns the last stage's content — the closest thing to a final recommendation in the brief. get_decision returns every completed stage.

Jobs

Anything that triggers LLM work (interview_next_question, synthesize_context, assess_decision, advance_stage, generate_framework_analysis) returns a job_id immediately rather than blocking. Poll get_job / GET /v1/jobs/{id} until status is completed or failed:

{
  "job_id": "…",
  "status": "completed",
  "output": { "stage_count": 4 },
  "error": null
}

This exists because LLM calls can take anywhere from a few seconds to a minute — long enough that an agent waiting synchronously will often time out and retry. Returning a job ID immediately sidesteps that, and pairing it with an Idempotency-Key (REST only — see Errors & Rate Limits) means a retried request replays the same job instead of starting a second one.

Credits

Every account has a monthly token allowance (plus any purchased top-ups) shared across the whole DeeCee.ai account — web app usage and agentic API usage draw from the same pool. Read-only calls (get_decision, list_frameworks, etc.) are free; anything that calls an LLM is metered by actual token usage. See Credits & Pricing for the exact rules.