ActionWeave Docs
ActionWeave is an AI workflow-orchestration platform. You connect your tools, define rules and guardrails, and an LLM orchestrator decides which skills and agents to run for every incoming event — while keeping you in control of anything sensitive.
Reasoning
An LLM plans each response under your guardrails.
Skills
Narrow playbooks keep the AI focused and safe.
Memory
pgvector memory makes every run smarter.
How it works
Every event runs the same loop. This predictability is what makes the platform auditable and safe.
An event happens
A new email, a webhook, a schedule, or an API call becomes a trigger.
The orchestrator decides
An LLM reads the event against your guidelines, guardrails, and memory, then picks the right playbook.
A skill runs its playbook
The matched skill exposes only the actions it is allowed to use.
Agents do the work
Specialized agents send emails, update CRMs, generate documents, or call connected systems.
Review and loop
Results persist to memory; sensitive actions route to you for approval.
Architecture
One orchestration loop, six layers. Each layer has a single job.
Triggers
Events from email, webhooks, schedules, voice, or the API.
Orchestrator
LLM reasons over context, guardrails, and memory.
Skills
Narrow playbooks with instructions + guardrails.
Capabilities
The allowlisted actions a skill may call.
Agents
Execute actions through real connectors.
Memory
PostgreSQL + pgvector feeds the next decision.
The loop
After agents act, results flow back into memory and the orchestrator decides whether more is needed. Skills can dispatch to other skills via aw.skill_dispatch, with depth, cycle, and fan-out guards preventing runaway loops.
Core concepts
Trigger
An inbound event (email, webhook, schedule, voice, or API call) that starts an execution.
Orchestrator
The LLM that interprets the trigger and plans the response within your guardrails.
Skill
A data-driven playbook that matches certain triggers and constrains the AI to a specific job and a fixed set of allowed actions.
Capability / Action
A single operation an agent can perform (send email, update CRM record, generate PDF). Skills allowlist which capabilities are available.
Agent
A specialized executor (Email, CRM, TMS, Billing, Documents, …) that carries out capabilities through real connectors.
Memory
Semantic store (PostgreSQL + pgvector) that records entities and execution results for cross-run context.
Tenant
An isolated customer boundary. Triggers, executions, and entities are scoped to a registered tenant.
Operating modes
Live
All actions enabled. Full production autonomy.
Internal-Only
External communications blocked; internal ops only.
Read-Only
Analysis only — reports what it would do without acting.
API reference
The Public API is served under /v1. Authenticate every request with an X-API-Key header. Partners can scope a call to a sub-account with X-Account-Id. API keys carry one of four scopes: full, trigger, read, or write.
# Fire a trigger via API
curl -X POST \
app.actionweave.ai/api/v1/trigger \
-H "X-API-Key: aw_live_xxx" \
-d {
"trigger_type": "email_received",
"source": "my_app",
"context": { ... }
}
Endpoints
/v1/triggerFire an event into the orchestrator.
/v1/executionsList executions; /v1/executions/{id} for detail.
/v1/monitorsHealth and run statistics.
/v1/agentsList agents; POST /v1/agents/{name}/execute to run one directly.
/v1/accountsCreate and manage partner sub-accounts.
/v1/connectorsList and manage OAuth connectors per account.
/v1/configUpdate context, instructions, and operating mode.
/v1/webhooksRegister webhook callbacks for async results.
/v1/usageUsage metrics per account / sub-account.
/v1/entitiesBrowse stored entities; intelligence routes add semantic search.
/v1/skills/catalogBrowse the skills catalog; register, enable, disable, and match-debug skills.
See the in-repo backend/docs/API_REFERENCE.md for full request/response schemas.
Ready to build?
See how ActionWeave orchestrates your workflows — in a live walkthrough.
Book a Demo