matic.shThe AI operating system for autonomous ventures
Define a goal, staff a team of AI agents, and let them do the work — on your filesystem, versioned in git, governed by policy.
Define a goal, staff a team of AI agents, and let them do the work — on your filesystem, versioned in git, governed by policy.
matic is an operating system for running structured teams of AI agents. You define a goal, connect tools and data, and operate agents that work together to achieve real business outcomes.
Instead of acting like a chatbot or a single-purpose workflow, matic takes inspiration from how human organisations work — structuring autonomous work around goals, teams, tasks, routines, memory, policies, and communication.
matic doesn't call AI models directly. It sits above agent runtimes like Claude Code, Codex CLI, and Gemini CLI, coordinating them through a normalised adapter layer. No lock-in at any level of the stack.
matic turns inbound signals into coordinated work, executes against defined outcomes, and compounds what it learns over time.
IngestChannel message→Webhook event→Routine emission→Signal
RouteSignal→Classify & interpret→Route to agent or group→Activation
PlanActivation→Decompose into tasks→Staff agents via probes→Work Items
ExecuteWork Items→Assemble context envelope→Agent runtime executes→Run
DeliverRun→Validate against contract→HITL checkpoint→Artifact
LearnArtifact→Observe outcome→Hypothesize improvements→Evolve probes & memory
The learning loop is not aspirational — it's a mandatory phase. After every engagement, agents enter a learning state where experience is consolidated, probes are refined, and institutional knowledge is captured as reusable archetypes.
Filesystem-first. All operational state — agents, projects, tasks, memory, policies — is stored as markdown and YAML files. No database required.
Git-native. Git is the persistence and collaboration layer. Every state change is a commit. Work is inspectable, versioned, diffable, and recoverable.
Agent-runtime agnostic. matic coordinates agent runtimes through adapters that normalise their interfaces. Swap runtimes without changing your org.
Channel-driven. Humans interact through channels — Slack, Telegram, WhatsApp, Discord, email, or the built-in terminal. Shell access is optional, not required.
Human control is first-class. Policies, guardrails, HITL checkpoints, and decision records ensure humans stay in the loop where it matters. Full audit trail by default.
| Primitive | Purpose |
|---|---|
| Org | The root operational context — a git repository containing everything |
| Charter | The org's governing objectives, constraints, and scope |
| Agent | A persistent autonomous actor with identity, memory, and evolving competence |
| Project | A bounded, goal-directed body of work |
| Team | A composition of agents assembled to execute a project |
| Signal | An inbound event from channels, webhooks, routines, or schedules |
| Work Item | A bounded deliverable with acceptance criteria and a floor/ceiling contract |
| Run | An execution record capturing what an agent did, when, and what it produced |
| Policy | A declarative rule defining what is permitted, with enforcement levels |
matic is designed for running entire business operations autonomously — not just answering questions or generating content:
| Section | What you'll find |
|---|---|
| Getting Started | Installation, first org, first agent, first run |
| Core Concepts | The mental model — scopes, primitives, work lifecycle |
| Agents | Identity, archetypes, probes, lifecycle, collaboration |
| Workflow | Signals, activation, work items, runs, delivery |
| Governance | Policies, HITL, decisions, audit, budgets |
| Platform | CLI, daemon, channels, runtimes, MCP, plugins |
| Reference | Schemas, configuration, filesystem layout, appendix |