Workflows: overview¶
Workflows let you automate multi-step processes that combine Autopoetic's chat and knowledge with your own agents and services — with peer review whose confidence Autopoetic computes for you.
A workflow has two layers, one vocabulary each:
- The engine runs templates you author: ordered steps that call out to external services over webhooks, wait for their agents to respond, branch on results, and loop until a condition is met. Template executions are called runs.
- The decision engine tracks how good each of your agents is at what it does — competencies per skill, updated by review outcomes — and decides, on every review round, whether the evidence is enough to trust an answer.
The engine orchestrates; the decision engine judges.
What you can build¶
- External agent orchestration. POST a work request to your own service and let the run wait — for seconds or days, at zero cost — until your agent calls back. Waiting is a first-class state, not a polling loop, and there is no deadline.
- Review workflows. Gate any step behind peer review: a producer submits work, your reviewer service picks an agent, and the run repeats review rounds until the decision engine's confidence crosses your threshold — or halts.
- Competency tracking. Every review outcome nudges each participant's competency scores; query them, their trends, and their history over time to decide who should review or produce next.
Where the API lives¶
Workflow endpoints are part of the platform API — the management surface that never runs inference — and are scoped under your project:
The full request/response schema — templates, runs, and the decision
engine's agents, work units, and adjustments — is on the live OpenAPI
schema page:
app.autopoetic.ai/schema
(raw JSON at /schema/openapi.json).
Project API keys authenticate; see
projects and teams.
In this section¶
- Writing templates — the template language: conditions, actions, and the review clause.
- Runs — starting, watching, waiting, and canceling.
- Review workflows — connecting your agents via webhooks and passphrases.
- The decision engine — agents, competencies, work units, and the confidence ladder.