Everything a local workbench should do

Local chat

A fast, familiar conversation surface in your browser, powered entirely by the runtime on your machine.

• Conversations stay in their workspace
• Choose providers per workspace or conversation
• Keep the browser as a simple, familiar control surface

Provider selection

Switch between OpenAI-compatible providers without changing the rest of the workspace. Hosted endpoints and local models can coexist.

• Configuration defines which providers load
• Local models are welcome
• Workspace policy keeps choices explicit

Durable memory

Facts, skills, and Mem0-backed memory persist per workspace, so useful context stays useful without becoming shared context.

• Memory remains scoped to the active workspace
• Skills are part of the composed runtime
• Compliance can audit each workspace on its own terms

Controlled file access

Local services expose safe file indexing and safe file access — governed, auditable, and never a free-for-all.

• Explicit file scope rather than broad machine access
• Local indexing services support retrieval
• Workspace configuration documents what is available

Markdown RAG

Workspace markdown and indexed files come back through retrieval layers, including contextual search.

• Persistent indexes stay local
• Retrieval is isolated by workspace
• Documentation and project files remain under your control

Expert panel

Select domain experts to answer independently, then synthesize their opinions into one decision with citations.

• Run legal, security, finance, compliance, and operations experts in parallel
• Choose the panel per question
• Receive one synthesized, cited answer