// concepts · positioning
Why Okami
Okami is not just another chat UI for code. The product is a verifiable, multi-channel, versionable harness that reduces dependence on a single frontier model.
Thesis
Okami turns agent behavior into operational contracts: tool registry, approval, sandbox, exit criteria, auditable skills, versioned memory and traces. The model still matters, but reliability no longer lives only inside the model.
Technical comparison
| dimension | Okami | typical tools |
|---|---|---|
| completion | task_complete is checked by exit criteria | often trusts final answer |
| skills | auditable files + scan + individual pages | less visible prompt/policy |
| channels | terminal, gateway, API, Telegram, Paperclip | usually IDE or terminal |
| model | swappable provider with capability profile | optimized around one provider |
| operations | events, replay, audit, retention and strict policy | logs are UI-first or scattered |
| design system | skill guides; ui_gate enforces | often textual instruction only |
Best fit
- Verifiable tasks outside the IDE: CI, cron, bot, API or Paperclip.
- Different models without rewriting workflows.
- Execution evidence instead of confident prose.
- Design-system and security policy enforced by contract.