captain-ai

When you are weighing a job offer, the real questions are rarely just is this company good. Captain AI turns four hard questions into a recoverable multi-agent workflow: do I actually fit this role, which claims have verifiable primary-source evidence behind them, are my hard constraints met, and what unknowns could change my decision. It starts from a confirmed and versioned job-seeking Profile, runs a bounded research pipeline through parallel Workers and an independent evidence Reviewer, and delivers a four-outcome decision report with every claim traced back to an accepted source. The final say always stays with you.\n\nThe core strength is dynamic-but-bounded multi-agent orchestration. A Planner decomposes open questions into one to six tightly scoped tasks; each Worker reads primary sources in minimal context and emits a structured evidence draft; a Reviewer that is blind to your Profile vets every draft against the original text and downgrades or rejects anything lacking direct support; a Context Selector recalls only the Profile items this specific judgment needs rather than dumping your full dossier. The state machine is owned by deterministic code with checkpoint recovery, so an interrupted run resumes exactly where it left off without re-parsing saved files or re-asking answered questions.\n\nIt is built for job seekers juggling multiple offers who want auditable analysis instead of pure gut feeling. Captain does not promise certainty or dispense motivational platitudes; it marks what it does not know, keeps unknowns honest out of any score, preserves your original wording, and never mutates a confirmed version without your explicit step. It is the friend in the storm who says look clearly and steer for yourself.