AI Threads
Location: /studio/[project]/ai
The AI Workbench is an agent-style chat workspace inside a project where you create story, evaluation, or general threads to develop your project with an AI. You can inject project context (worldview, characters, scenario versions, shot pipelines) as selectable chips, and replies stream in. Turning on Generate mode (✦) plans shots from your intent and renders images directly into the reply, and any existing assistant reply can be turned into images with one button. Threads persist per project.
- 1Create a thread with the +Story / +Evaluation / +General buttons.
- 2In the context area, toggle chips (worldview, characters, scenario, etc.) to inject project data (select all/clear).
- 3Confirm the provider (BYOK key or local CLI agent) and type your request.
- 4Send (or Cmd+Enter) to get a streaming reply.
- 5For images, enable the ✦ Generate toggle, pick a genre/purpose and send, or click 'Generate from reply' on an assistant message.
Chat: BYOK providers (OpenAI / Anthropic Claude / Google Gemini) or local CLI agents (Codex, Claude Code). Image generation: gpt-image via god-tibo-imagen (keyless, /api/owntent/imagen).
AI connections (Codex/Claude Code) and MCP servers are managed in Integrations (/integrations), not in the Workbench. If no provider is available, add a BYOK key or enable a CLI agent in Settings. The first message auto-names the thread.