Configuration Reference

AOF uses three layers of configuration: the org chart (agent topology and routing), the AOF config (runtime behavior), and the OpenClaw plugin wiring (gateway integration). This document covers all three.

---

Org Chart (org-chart.yaml)

The org chart is a YAML file that defines agents, teams, organizational units, routing rules, and memory pools. It is the single source of truth for “who can do what” in your AOF deployment.

All org chart changes should go through aof config set or aof config apply for validation and atomic writes. Source: src/schemas/org-chart.ts.

Top-level Fields

Field	Type	Required	Default	Description
schemaVersion	1	Yes	—	Schema version, must be 1
template	string	No	—	Org template name (e.g., "swe-team", "ops-team")
agents	Agent[]	Yes	—	List of agent definitions
teams	Team[]	No	[]	Team definitions (legacy; use orgUnits for P1.1+)
routing	RoutingRule[]	No	[]	Tag/priority-based routing rules (legacy; use relationships for P1.1+)
orgUnits	OrgUnit[]	No	[]	Organizational units (department, team, squad)
groups	OrgGroup[]	No	[]	Cross-cutting agent groups
memberships	OrgMembership[]	No	[]	Agent-to-org-unit memberships
relationships	OrgRelationship[]	No	[]	Inter-agent relationships (escalation, delegation, etc.)
defaults	OrgDefaults	No	—	Default policies and capabilities
memoryPools	MemoryPools	No	—	Memory V2 pool definitions
memoryCuration	MemoryCuration	No	—	Memory curation configuration
roles	Record<string, RoleMapping>	No	—	Role-based agent mapping for DAG workflows
metadata	Record<string, unknown>	No	{}	Arbitrary metadata

Agent Definition (agents[])

Each agent represents a single autonomous entity that can receive and execute tasks.

Field	Type	Required	Default	Description
id	string	Yes	—	Unique agent ID (must match OpenClaw agent ID)
openclawAgentId	string	No	—	OpenClaw agent ID (e.g., "agent:main:main"). Used for drift detection
name	string	Yes	—	Human-readable display name
description	string	No	—	Role description
team	string	No	—	Team membership (legacy, use memberships for P1.1+)
reportsTo	string	No	—	Reports-to agent ID (legacy)
canDelegate	boolean	No	false	Whether this agent can delegate tasks to others
capabilities	AgentCapabilities	No	{}	Capability tags and concurrency settings
comms	AgentComms	No	{}	Communication/dispatch preferences
policies	OrgPolicies	No	—	Agent-specific policy overrides
active	boolean	No	true	Whether the dispatcher considers this agent

Agent Capabilities

Field	Type	Default	Description
tags	string[]	[]	Capability tags used for routing (e.g., ["typescript", "backend"])
concurrency	number	1	Maximum concurrent tasks this agent can handle
model	string	—	Model assigned to this agent (informational)
provider	string	—	Provider type (informational, for cost tracking)

Agent Communication Preferences

Field	Type	Default	Description
preferred	"spawn" | "send" | "cli"	"send"	Preferred dispatch method
sessionKey	string	—	Session key for sessions_send dispatch
fallbacks	("spawn" | "send" | "cli")[]	["send", "cli"]	Fallback methods in priority order

Team Definition (teams[])

Field	Type	Required	Default	Description
id	string	Yes	—	Unique team identifier
name	string	Yes	—	Human-readable team name
description	string	No	—	Team description
lead	string	No	—	Lead agent ID
orchestrator	string	No	—	Orchestrator agent ID (manages tasks and reviews)
technicalLead	string	No	—	Technical lead agent ID
murmur	MurmurConfig	No	—	Orchestration review configuration
dispatch	TeamDispatchConfig	No	—	Per-team dispatch throttling overrides

Team Dispatch Throttling

Field	Type	Default	Description
maxConcurrent	number	—	Maximum concurrent dispatches for this team (overrides global)
minIntervalMs	number	—	Minimum interval between dispatches in ms (overrides global)

Murmur Orchestration

Murmur triggers orchestration reviews when specific conditions are met.

Field	Type	Description
triggers	MurmurTrigger[]	At least one trigger condition
context	string[]	Context to inject: "vision", "roadmap", "taskSummary"

Trigger kinds: "queueEmpty", "completionBatch", "interval", "failureBatch"

Routing Rules (routing[])

Routing rules determine which agent receives a task based on tags and priority.

Field	Type	Default	Description
matchTags	string[]	[]	Match tasks with these capability tags
matchPriority	string[]	[]	Match tasks with these priorities
targetRole	string	—	Route to agents with this role
targetTeam	string	—	Route to this team
targetAgent	string	—	Route to this specific agent
weight	number	100	Rule priority (lower = evaluated first)

Organizational Units (orgUnits[])

P1.1 extension for tree-structured org hierarchy.

Field	Type	Required	Default	Description
id	string	Yes	—	Unique org unit ID
name	string	Yes	—	Human-readable name
type	string	Yes	—	Unit type (department, team, squad, etc.)
parentId	string	No	—	Parent org unit ID (for tree structure)
description	string	No	—	Description
leadId	string	No	—	Lead agent ID
active	boolean	No	true	Whether this unit is active
metadata	Record<string, unknown>	No	{}	Arbitrary metadata

Relationships (relationships[])

Inter-agent relationships for escalation, delegation, and consultation.

Field	Type	Required	Default	Description
fromAgentId	string	Yes	—	Source agent ID
toAgentId	string	Yes	—	Target agent ID
type	enum	Yes	—	"escalates_to", "delegates_to", "consults_with", or "reports_to"
active	boolean	No	true	Whether this relationship is active
metadata	Record<string, unknown>	No	{}	Arbitrary metadata

Role Mapping (roles)

Maps abstract roles to concrete agents for DAG workflows.

Field	Type	Required	Default	Description
agents	string[]	Yes	—	Agent IDs that can fulfill this role (at least one)
description	string	No	—	Human-readable description of responsibilities
requireHuman	boolean	No	—	Whether this role requires human involvement

Memory Pools (memoryPools)

Tiered memory pool definitions for the Memory V2 system.

Field	Type	Required	Default	Description
hot	MemoryPoolHot	Yes	—	Hot tier (always indexed)
hot.path	string	Yes	—	Path to hot pool root
hot.description	string	No	—	Pool description
hot.agents	string[]	No	—	Explicit agent list (all agents if omitted)
warm	MemoryPoolWarm[]	Yes	—	Warm tier pools (role-scoped)
warm[].id	string	Yes	—	Unique pool ID
warm[].path	string	Yes	—	Path to warm pool root
warm[].roles	string[]	Yes	—	Role/agent patterns that include this pool
cold	string[]	Yes	—	Cold tier paths
adapter	"filesystem" | "lancedb"	No	"filesystem"	Memory retrieval adapter

Policies

Policies can be set at org-chart defaults level or per-agent.

Memory Policy

Field	Type	Default	Description
scope	string[]	—	Memory scope paths (e.g., ["org/engineering", "shared/docs"])
tiers	("hot" | "warm" | "cold")[]	—	Allowed memory tiers
readOnly	boolean	false	Read-only access

Tasking Policy

Field	Type	Default	Description
maxConcurrent	number	1	Maximum concurrent tasks
allowSelfAssign	boolean	false	Whether agent can self-assign tasks
requiresReview	boolean	false	Whether tasks require review
allowedPriorities	string[]	—	Allowed task priorities

Communication Policy

Field	Type	Default	Description
allowedChannels	string[]	—	Allowed communication channels
requiresApproval	boolean	false	Whether communication requires approval
restrictedAgents	string[]	[]	Agent IDs that cannot be communicated with

Context Budget Policy

Prevents context rot by limiting how much context an agent receives.

Field	Type	Description
target	number	Target budget in characters (ideal context size)
warn	number	Warning threshold in characters
critical	number	Critical threshold in characters (must truncate)

Example: Minimal Org Chart

schemaVersion: 1
agents:
  - id: main
    name: Main Agent
    capabilities:
      tags: [general]
      concurrency: 1
    comms:
      preferred: send
teams: []
routing: []

Example: Team with Roles

schemaVersion: 1
agents:
  - id: swe-backend
    name: Backend Engineer
    capabilities:
      tags: [typescript, nodejs, apis]
      concurrency: 2
    comms:
      preferred: send
  - id: swe-frontend
    name: Frontend Engineer
    capabilities:
      tags: [react, css, ui]
      concurrency: 1
    comms:
      preferred: spawn
  - id: swe-lead
    name: Engineering Lead
    capabilities:
      tags: [review, architecture]
      concurrency: 1
    canDelegate: true

teams:
  - id: engineering
    name: Engineering
    lead: swe-lead

roles:
  backend:
    agents: [swe-backend]
    description: Backend implementation
  frontend:
    agents: [swe-frontend]
    description: Frontend implementation
  reviewer:
    agents: [swe-lead]
    description: Code review

routing:
  - matchTags: [backend, api]
    targetRole: backend
    weight: 10
  - matchTags: [frontend, ui]
    targetRole: frontend
    weight: 10

Example: Skill-based Routing

schemaVersion: 1
agents:
  - id: agent-alpha
    name: Alpha
    capabilities:
      tags: [python, ml, data]
      concurrency: 3
  - id: agent-beta
    name: Beta
    capabilities:
      tags: [typescript, nodejs, devops]
      concurrency: 2

routing:
  - matchTags: [ml, data]
    targetAgent: agent-alpha
    weight: 1
  - matchTags: [devops]
    targetAgent: agent-beta
    weight: 1
  - matchPriority: [critical]
    targetAgent: agent-alpha
    weight: 0    # highest priority rule

---

AOF Configuration

AOF runtime configuration controls the scheduler, event logging, metrics, and communication behavior. Source: src/schemas/config.ts.

Top-level Config

Field	Type	Default	Description
schemaVersion	1	—	Schema version, must be 1
dataDir	string	"~/.aof/data"	Root data directory for AOF runtime data
orgChartPath	string	"org-chart.yaml"	Path to org chart YAML file
vaultRoot	string	—	Root directory for vault (Projects/, Resources/)
dispatcher	DispatcherConfig	{}	Scheduler/dispatch settings
daemon	DaemonConfig	{}	Daemon runtime settings
metrics	MetricsConfig	{}	Prometheus metrics settings
eventLog	EventLogConfig	{}	Event logging settings
comms	CommsConfig	{}	Communication fallback settings
metadata	Record<string, unknown>	{}	Arbitrary metadata

Dispatcher Configuration

Controls how the scheduler scans for tasks and dispatches them.

Field	Type	Default	Description
intervalMs	number	120000 (2 min)	How often the dispatcher scans for pending tasks
defaultLeaseTtlMs	number	600000 (10 min)	Default lease TTL for task assignments
maxLeaseRenewals	number	3	Max lease renewals before force-expiry
dryRun	boolean	false	Log decisions without dispatching
maxConcurrentDispatches	number	3	Maximum concurrent dispatches globally
minDispatchIntervalMs	number	0	Minimum interval between dispatches (0 = disabled)
maxDispatchesPerPoll	number	10	Maximum dispatches per poll cycle

Daemon Configuration

Controls the aof-daemon runtime. Source: src/config/registry.ts.

Field	Type	Default	Description
pollIntervalMs	number	30000	Poll interval for the scheduler loop (ms)
socketPath	string	<dataDir>/daemon.sock	Override path for the Unix-domain IPC socket. Created with mode 0600.
mode	"plugin-bridge" | "standalone"	"standalone"	Dispatch mode selector. plugin-bridge expects an OpenClaw plugin to attach via long-poll and delegates agent spawns to it; standalone falls through to the HTTP gateway adapter. Adapter selection also happens at dispatch time based on whether a plugin is currently attached — the mode value controls the “no plugin attached → hold” behavior (D-12).

Metrics Configuration

Field	Type	Default	Description
enabled	boolean	true	Enable Prometheus metrics export
port	number	9101	Port for the metrics HTTP server
path	string	"/metrics"	Metrics endpoint path

Event Log Configuration

Field	Type	Default	Description
enabled	boolean	true	Enable event logging
maxEventsPerFile	number	10000	Max events per log file before rotation
maxFiles	number	30	Max total log files to retain

Communication Configuration

Controls how the dispatcher communicates with agents.

Field	Type	Default	Description
methodPriority	("spawn" | "send" | "cli")[]	["send", "spawn", "cli"]	Default dispatch method priority
spawnTimeoutMs	number	30000	Timeout for spawn attempts
sendTimeoutMs	number	60000	Timeout for send attempts
cliTimeoutMs	number	120000	Timeout for CLI attempts

---

OpenClaw Plugin Wiring

AOF registers as an OpenClaw plugin via openclaw.plugin.json. As of v1.15, the plugin is a thin bridge: it forwards tool invocations to the daemon over daemon.sock and handles agent-spawn callbacks from the daemon via a long-poll. It no longer owns an in-process AOFService — that lives in the daemon.

This section covers the plugin manifest and the gateway configuration needed to activate AOF.

Plugin Manifest (openclaw.plugin.json)

The plugin manifest declares AOF’s identity and configuration schema. Source: openclaw.plugin.json.

Field	Value	Description
id	"aof"	Plugin identifier
name	"AOF -- Agentic Ops Fabric"	Display name
version	"0.1.0"	Plugin version
kind	"memory"	Plugin kind
main	"dist/plugin.js"	Entry point
configSchema	object	JSON Schema for plugin config (see below)

Plugin Config Schema

These fields go in the plugins.aof.config section of your openclaw.json:

Field	Type	Default	Description
dataDir	string	"~/.aof/data"	AOF state directory (must match the daemon’s)
pollIntervalMs	number	30000	Scheduler poll interval in milliseconds
defaultLeaseTtlMs	number	300000	Default task lease TTL in milliseconds
dryRun	boolean	false	If true, scheduler observes but does not dispatch
gatewayUrl	string	auto-detected	OpenClaw gateway URL for HTTP dispatch
gatewayToken	string	auto-detected	Gateway authentication token
heartbeatTtlMs	number	300000	Heartbeat TTL in milliseconds
maxConcurrentDispatches	number	3	Maximum concurrent dispatched tasks (1-50)
modules	object	—	Module enable/disable flags
modules.memory.enabled	boolean	true	Enable memory module
modules.dispatch.enabled	boolean	true	Enable dispatch module
modules.murmur.enabled	boolean	true	Enable murmur orchestration module
modules.linter.enabled	boolean	true	Enable linter module
memory	object	—	Memory subsystem configuration
memory.embedding.provider	"openai" | "ollama"	—	Embedding provider
memory.embedding.model	string	—	Embedding model name
memory.embedding.baseUrl	string	—	Embedding API base URL
memory.embedding.apiKey	string	—	Embedding API key
memory.embedding.dimensions	number	—	Embedding dimensions
memory.search.hybridEnabled	boolean	true	Enable hybrid search (vector + BM25)
memory.search.vectorWeight	number	0.7	Vector search weight (0-1)
memory.search.bm25Weight	number	0.3	BM25 search weight (0-1)
memory.search.maxResults	number	10	Maximum search results
memory.search.tierBoost.hot	number	1.0	Score boost for hot tier results
memory.search.tierBoost.warm	number	0.8	Score boost for warm tier results
memory.search.tierBoost.cold	number	0.5	Score boost for cold tier results
memory.indexPaths	string[]	—	Additional paths to index

Gateway Configuration Example

To enable AOF in your OpenClaw gateway, add it to the plugins section of ~/.openclaw/openclaw.json:

{
  "plugins": {
    "aof": {
      "enabled": true,
      "config": {
        "dataDir": "~/.aof/data",
        "pollIntervalMs": 30000,
        "maxConcurrentDispatches": 3,
        "modules": {
          "memory": { "enabled": true },
          "dispatch": { "enabled": true }
        },
        "memory": {
          "embedding": {
            "provider": "openai",
            "model": "text-embedding-3-small",
            "dimensions": 1536
          },
          "search": {
            "hybridEnabled": true,
            "vectorWeight": 0.7,
            "bm25Weight": 0.3
          }
        }
      }
    }
  }
}

Note: gatewayUrl and gatewayToken are auto-detected from the OpenClaw runtime context. Only set them explicitly if auto-detection fails.

Managing Config via CLI

AOF provides CLI commands for org chart configuration management:

# Get a config value (dot-notation)
aof config get agents.swe-backend.active

# Set a config value (validates + atomic write)
aof config set agents.swe-backend.active true

# Preview a change without applying
aof config set agents.swe-backend.active false --dry-run

# Validate the entire org chart
aof org validate

All changes through aof config set are validated against the Zod schema and checked for referential integrity before being written atomically (write to temp file, validate, rename).