Lists which AI providers (claude, openai, gemini) currently have API keys configured and are usable in the other tools.

Lists the specialists registered for run_orchestrator: id, role, which provider currently performs the role, and whether the specialist gathers external evidence. Pass these ids in run_orchestrator's workers array instead of redeclaring each specialist.

Lists what each model can do as wired up in this codebase — not what its vendor advertises. Use it to see which models support grounded retrieval, and which have verified pricing.

Runs a sequential pipeline where each step is handled by a chosen AI provider, and each step's output feeds into the next step's input (accessible via {{input}} in promptTemplate).

A Chief provider classifies the task's complexity, picks the minimum sufficient set of specialists, gives each one coherent mission, then a synthesizer merges their results into one final answer. Specialist count is capped by complexity (simple 1, normal 3, deep 4). Returns a run report: SUCCESS/DEGRADED/FAILED plus an evidence label, and a deep answer with no evidence behind it is marked UNVERIFIED in the output itself.

Multiple AI providers independently answer the same question, then critique each other's answers over one or more rounds, and finally a judge provider synthesizes or selects the best final answer.