Flows Overview

Flows are deterministic YAML workflows that capture proven agent behavior as repeatable, schedulable automation. Where chat is exploratory and open-ended, a flow is a fixed node graph executed by Astonish's agent state machine — producing the same result every time given the same inputs.

What Is a Flow?

A flow is a directed graph of nodes (actions) connected by edges (transitions). Each node performs a single operation — calling an LLM, invoking a tool, evaluating a condition, or collecting input — and passes state to the next node in the graph.

name: summarize-pr
description: Summarize a GitHub pull request and post to Slack
params:
  - name: pr_url
    type: string
    required: true

nodes:
  - id: fetch_pr
    type: tool
    tool: github_get_pr
    args:
      url: "{{pr_url}}"

  - id: summarize
    type: llm
    prompt: "Summarize this PR for a team standup: {{fetch_pr.output}}"

  - id: post
    type: tool
    tool: slack_post
    args:
      channel: "#engineering"
      text: "{{summarize.output}}"

edges:
  - from: fetch_pr
    to: summarize
  - from: summarize
    to: post

How Flows Differ from Chat

Aspect	Chat	Flow
Execution	Non-deterministic, exploratory	Deterministic, repeatable
Versioning	Session transcript	YAML in git or database
Scheduling	Manual invocation	Cron, webhook, or event trigger
Sharing	Share a transcript	Share executable automation
Parameters	Freeform prompts	Declared typed inputs

The Flow Lifecycle

Flows follow a natural progression from discovery to production:

1. Chat — Solve a problem interactively with the agent.
2. Distill — Run /distill to extract a parameterized flow from the session. See Flow Distillation.
3. Edit — Refine the YAML directly or through the Studio visual editor.
4. Schedule — Attach a cron expression, webhook trigger, or event subscription.
5. Share — Publish to your team via Studio or distribute via Taps.

Storage & Multi-Tenancy

Flows are stored in the platform database with full multi-tenancy:

• Each flow belongs to a user or team.
• Flows can be private, team-shared, or published to the organization.
• Version history is maintained — roll back to any previous revision.
• Execution logs are retained for audit and debugging.

Running a Flow

# Run from CLI
astonish flows run summarize-pr -p pr_url="https://github.com/org/repo/pull/42"

In Studio, navigate to the Flows tab, select a flow, click Run, and fill in the parameters visually.

Flows can also be triggered programmatically via the REST API, making them composable building blocks for larger automation pipelines.

Next Steps

• Flow Distillation — Create flows from chat sessions
• YAML Reference — Full schema documentation
• Nodes, Edges & State — Deep dive into flow mechanics
• Taps & Flow Store — Community flow repositories