One prompt, sent in parallel to whichever AI Providers you check below — responses shown side by side for comparison. Follow-up questions build on the conversation below until you start a new one.
Plain conversation with one model at a time. Follow-ups build on this thread until you start a new one.
Just ask — a target (an AI Provider or a logged-in Claude/Codex slot) is picked automatically per question. Coding-shaped prompts route to a CLI slot; everything else routes to your fastest-performing provider.
Several models answer independently, then a judge model reads every answer and synthesizes one verdict — agreement, conflict, and which reasoning holds up. Slower than Council, meant for a real decision rather than a quick comparison. Each run is a one-shot artifact, not an ongoing conversation.
A real working directory the agent can read/write files in across turns. Switch provider mid-conversation — context carries over even across engines.
Log in up to 6 accounts once each. Each connected project picks its own active slot independently — switching one project's active account never affects another's.
1. A sign-in tab just opened (or click here) — sign in with the account for .
2. Paste the code shown on that page below.
Device-code login — a URL and short code, approved in any browser. No code to paste back here; the login finishes on its own once approved.
Each project (octozen, Bnlzen, …) gets its own token. Pick a slot here to fix it for that project (used whenever its API call omits slot), or leave on Auto and let the project pass slot itself — an explicit slot in a call always wins. "last used" is a best-effort mirror of the slot it most recently called with. The AI provider column does the same for /v1/chat (OpenRouter, Qwen, etc. from the AI Providers list below): pick a default, and use Allowed to restrict which providers that project may call (all allowed unless you untick some).
| Name | Token | Claude slot (assigned) | Codex slot (assigned) | AI provider (assigned / allowed) | Last seen |
|---|
Any OpenAI-Chat-Completions-compatible endpoint (NVIDIA NIM, Groq, Gemini's OpenAI-compat endpoint, a local Qwen3 Coder server, …). Used by Council above and handed out to connected projects. No "Claude (built-in)" entry — the CLI accounts above already cover Claude.
| Name | Base URL | Model | API key |
|---|
One key, 300+ models — search OpenRouter's live catalog and add any model as its own provider (same as above, just without typing an exact model slug by hand).