Comparison
Shared heritage. A different architecture bet.
CareerOS owes a real debt to Career Ops. The comparison below is about design choices, not a ranking — both are open-source, MIT-licensed, and local-first.
Attribution
The Final Evaluation rubric and matching methodology in CareerOS are adapted from Career Ops. CareerOS deliberately diverges on architecture (host-CLI-driven, not a standalone bot), output format (structured JSON, not a long report for every job), and cost model (gate before evaluate, cache everything).
| Dimension | Career Ops | CareerOS |
|---|---|---|
| Runtime model | Mode files run inside your AI coding CLI, which drives the whole flow. | A Python package (careeros) does the deterministic work; a host coding CLI performs only the reasoning steps. |
| Where AI is spent | The evaluation and tailoring are model-driven, per its A–F scoring across weighted dimensions. | AI is limited to a cheap gate and one structured evaluation per surviving job; discovery, filtering, and reporting spend no tokens. |
| Evaluation output | An A–F grade with tailored, ATS-oriented CV documents. | A compact JSON judgment (score, recommendation, strengths, weaknesses, fit paragraph) that every artifact reuses. |
| Cost model | Processes offers, including in batch, to evaluate many in parallel. | Gate before evaluate, plus fingerprinted caching keyed on job, profile version, and prompt version — an unchanged re-run costs zero calls. |
| Where results land | A tracked single source of truth with integrity checks. | A Google Sheet (newest on top) with a hand-editable status column, plus local run files as the source of truth. |
| Truthfulness enforcement | Generates tailored CVs customized per job description. | A deterministic verify step refuses to cache a resume whose lines don't verbatim-match your profile; a lint enforces voice. |
Career Ops descriptions reflect its own public documentation and may change as that project evolves. If anything here is out of date, the correction belongs upstream — please open an issue.