Facio is a Docker-first runtime for long-running AI work. It keeps session state, memory, tools, channels, and Placet review flows together so agents can continue where boundaries are clear and return to a human before sensitive actions.
Models · channels · skills · browser · approvals
What Facio actually does
Facio keeps state, resumes after interruptions, asks for authority at the right moment, and records what happened. It is designed for teams that need AI help without creating a silent data leak or an uncontrolled attack surface.
Facio pauses for human review when work touches credentials, production, files, or external systems.
Preferences, durable facts, known issues, and project context stay available after the chat window is gone.
Secret redaction, isolated credentials, policy-gated tools, and sandboxed execution keep scope explicit.
Actions, approvals, model choices, and outcomes stay visible enough for security and operations teams to trust.
Text, attachments, vision, transcription, image generation, and video generation run in one agent.
Switch models, providers, channels, and tools while work is running instead of redeploying the stack.
Extend at runtime
No rebuilds, no restarts. Add capabilities the moment the work needs them.
Live runtime
Facio routes work while it runs
Connect MCP servers over stdio or SSE, hot-install workspace skills, and capture OAuth or credentials only when the task needs them.
Fetch pages, run search, inspect web data, and keep every external step approval-gated and visible to the operator.
Execute commands with policy gates, redaction, seccomp/bwrap isolation, and a clear record of what changed.
Transcribe audio, read images, generate images or video, and route the work through the model best suited for the moment.
Traceable autonomy
Business teams worry about uncontrolled AI for good reasons: data exposure, hidden tool calls, unclear authority, and fragile handoffs. Facio makes the agent useful while keeping sensitive decisions reviewable and recoverable.
Audit trail
Plays well with everything
Switch live between commercial and local models, or run your own OpenAI-compatible endpoint.
Operate Facio from Placet, your team chat, your inbox, the terminal, or any A2A 1.0 client.
Ready in seconds
One command brings up the agent runtime, the Placet operator dashboard, and the HITL chat, all on your own infrastructure.