Overview

Facio is a long-running, human-in-the-loop runtime for reliable AI work. It keeps useful autonomy inside clear boundaries: the agent can continue work, use tools, remember context, and route through channels, while humans stay in control at risky or unclear moments.

The docs are written for people evaluating or operating Facio. They explain what happens at a practical level without exposing prompt internals, private thresholds, checkpoint keys, or low-level wire formats.

Facio operating model

Observe · Ask · Act · Record · Resume

A persistent agent runtime with Placet-first review flows, session state, memory, tools, channels, MCP, audit records, and Docker-first deployment.

01Getting StartedInstall with Docker, open Placet, and run a safe first task.02Runtime

How sessions, approvals, tools, memory, and recovery fit together.

03API Gateway

OpenAI-compatible HTTP, management API, Placet callbacks, and A2A.

04Python SDKEmbed a direct Facio agent run from a Python process.05ProvidersConfigure hosted, local, gateway, and OAuth model providers.06MCP

Add MCP servers, link credentials, and discover server listings.

07Channels

Placet, Telegram, Slack, Discord, Email, Teams, CLI, HTTP, and A2A.

08ConfigurationEnvironment variables, runtime settings, and credentials.09IntegrationsHow the integration pieces fit together at a product level.10Security

The public security model: sandboxing, redaction, allow lists, and review gates.

11Operations

Run, scale, update, inspect, and maintain a Facio environment.

Start

Quickstart

Runtime model

observe · ask · act · record · resume

Security

Integrations

Operations

What Facio is

Facio is the runtime layer around agent work. It can call tools, use models and providers, keep state across long tasks, route through supported channels, and pause when a human needs to approve a risky step.

The main product idea is simple: let the agent continue where automation is safe, but make authority, data access, side effects, and review visible. Placet is the default operator surface because structured approvals, file-heavy review, streaming progress, and management workflows fit this operating model better than plain chat.

Facio is biased toward work that lasts longer than one chat turn: research, operational follow-up, file-heavy review, multi-step execution, recovery after interruptions, and workflows where auditability matters.

Operating model

Facio combines five principles in one runtime:

Principle	Meaning
Observe	Keep enough session, memory, history, and runtime context to avoid restarting from zero.
Ask	Return to a human when risk, ambiguity, authority, or external side effects are unclear.
Act	Use models, tools, scripts, channels, browser work, and MCP only inside configured boundaries.
Record	Keep decisions, tool use, approvals, errors, and outcomes understandable after the task.
Resume	Continue after interruptions, approvals, retries, and delayed follow-ups with the right context.

In practice, this means Facio assembles context, sessions, memory, skills, tools, channel metadata, provider routing, and review checkpoints around each turn. The point is not the internal wiring; it is that Facio treats reliability as runtime behavior rather than a prompt convention.

When Facio fits

Facio is a good fit when a team needs one or more of these:

Long-running agent sessions that can resume after interruptions.
Human review before external, irreversible, or sensitive actions.
Audit trails for decisions, tools, approvals, and outcomes.
A Docker-first deployment that can run on controlled infrastructure.
Runtime integrations with models, channels, skills, scripts, browser work, media tools, and MCP servers.
A primary Placet workflow with secondary transports when chat, CLI, A2A, or OpenAI-compatible access is useful.

What Facio is not

Facio is not a thin chat UI around one model. It is also not meant to hide sensitive operations behind autonomous behavior. The runtime is most useful when the team wants the agent to work independently within clear boundaries and return to humans at authority boundaries.

Docs map

Page	Use it for
Getting Started	First local setup, first task, helper commands, and common setup issues.
Runtime	How a request moves through context, model routing, tools, HITL, memory, audit, and resume.
API Gateway	HTTP surfaces, bearer-token management API, OpenAI-compatible calls, and A2A.
Providers	Hosted, local, gateway, and OAuth provider setup.
MCP	Runtime MCP servers, agent-managed setup, credentials, and listings.
Channels	Placet, chat channels, email, Teams, CLI, HTTP, and A2A.
Configuration	Environment variables, config areas, credentials, and runtime settings.
Commands	Quickstart helper, CLI, slash commands, and management command endpoints.
Integrations	Product-level map of providers, Placet, channels, MCP, skills, scripts, browser work, and credentials.
Security	Public security model, sandboxing, secrets, guardrails, allow lists, and operator checklist.
Operations	Day-to-day commands, scaling, logs, updates, backups, and production-readiness checks.

Content boundaries

These docs intentionally avoid copying internal thresholds, prompt text, retry sequences, checkpoint keys, fallback strings, or protocol details. The public goal is to explain how operators should reason about Facio, not to expose a rebuild recipe for the runtime internals.

Next step

Read Getting Started to install the local Docker-first setup and run a first low-risk task.