aof

AOF - Agentic Ops Fabric

AOF lets you run teams of AI agents the way you'd run teams of people. Define your org structure, assign roles, and set up workflows. AOF handles the rest: routing tasks, enforcing review stages, recovering from crashes, and making sure nothing falls through the cracks.

It works for any domain where agents collaborate - software engineering, RevOps, sales, marketing, research, whatever. If your process has steps, AOF can govern it.

---

What It Does

• Org-chart-driven orchestration. Model agents, teams, and hierarchies in a YAML org chart. Work gets routed by capability, team, and priority.
• Workflow DAGs. Define multi-stage pipelines with review loops, conditional branches, and parallel fan-out. Agents follow the process - no skipping steps, no dropped handoffs.
• Shared memory. Agents build on each other's knowledge through semantic memory with HNSW vector search and tiered curation.
• Works for any domain. SWE (implement → review → QA → deploy), RevOps (qualify → enrich → handoff), sales pipelines, research workflows. If it has a process, it fits.
• Nothing gets dropped. Filesystem-first task store with atomic state transitions, lease-based locking, and deadletter recovery. Tasks survive crashes, restarts, and API failures.

---

Quick Start

Prerequisites

• Node.js >= 22 (LTS recommended)
• OpenClaw gateway running (openclaw.dev)

Install

curl -fsSL https://raw.githubusercontent.com/d0labs/aof/main/scripts/install.sh | sh

The installer scaffolds your data directory and installs the aof-daemon user service (launchd on macOS, systemd on Linux). The daemon is the single scheduler/IPC authority; the OpenClaw plugin connects to it over a Unix socket at ~/.aof/data/daemon.sock.

Set up and run

aof init              # Configure OpenClaw integration
aof daemon status     # Verify the daemon is running (installed by the installer)
aof org show          # List available agents
aof task create "My first task" --agent <your-agent-id>

See the Getting Started Guide for a complete zero-to-working walkthrough.

---

Upgrading

Two paths exist on paper:

aof update                                                                            # in-CLI updater
curl -fsSL https://raw.githubusercontent.com/d0labs/aof/main/scripts/install.sh | sh  # re-run the installer

Both preserve your data directory (~/.aof/data) and the daemon service. aof update also writes a backup so it can roll back with aof update --rollback --backup <path>.

Per-release notes live on GitHub Releases.

---

Key Features

| Feature | Description | Docs | |---------|-------------|------| | Org chart governance | YAML org charts define agents, teams, routing rules, and memory scopes | Configuration | | DAG workflows | Multi-stage pipelines with rejection loops and parallel fan-out | Workflow DAGs | | Protocol system | Typed inter-agent messages: handoff, resume, status update, completion | Protocols | | Semantic memory | HNSW vector index with hybrid search and tiered curation | Memory | | Recovery-first | Deadletter queue, task resurrection, lease expiration, drift detection | Recovery | | Observability | Prometheus metrics, JSONL events, Kanban board | Event Logs |

---

Documentation

For Users

• Getting Started - Install, configure, and orchestrate your first agent team
• Configuration Reference - Org chart schema, AOF config, OpenClaw plugin wiring
• CLI Reference - Complete command reference (auto-generated)
• Documentation Index - Full user and developer documentation

For Contributors

• Architecture Overview - System diagram, subsystem descriptions, key interfaces
• Dev Workflow - Development setup and fast-feedback loop

---

License

MIT - see LICENSE.