flowr 1.2.1

A declarative, validatable YAML format for non-deterministic state machine workflows.

Read the specification →

The specification

flowr defines what a workflow is — its states, transitions, guard conditions, subflows — not what it does. No execution engine. No side effects. No opinions about retries, timeouts, or error handling. A YAML file declares structure. A validator checks integrity. Tools query, track, and visualise. The format is the foundation.

What the specification covers:

• States with unique ids, per-state attributes, and transition mappings
• Transitions that resolve to state ids or declared exit names
• Guard conditions gated by evidence-based expressions using 6 operators (==, !=, >=, <=, >, <)
• Named condition groups reusable across transitions on the same state
• Subflows with call-stack semantics — push on entry, pop on exit
• Within-flow cycles for iterative workflows
• Validation rules — structural integrity checks independent of any runtime

What the specification does not cover:

• Execution engines, side-effect hooks, retry logic, timeout handling
• Parallel (fork-join) states
• Orchestration, scheduling, or event dispatch

→ Full specification with examples and visual diagrams

The reference implementation

This repository contains a Python reference implementation — a CLI and library that validates, queries, and tracks flow definitions conforming to the specification. The specification is the contract; this code is one tool that honours it.

flowr validate deploy.yaml          →  valid: True
flowr states deploy.yaml            →  prepare, execute, review
flowr next deploy.yaml review       →  approve → deployed [blocked]  need: score=>=80
                                      →  reject → failed
flowr transition deploy.yaml review approve --evidence score=85
                                      →  from: review, to: deployed
flowr session init deploy-flow       →  session created at state: prepare
flowr --session transition approve  →  from: prepare, to: review
flowr export deploy.yaml --format mermaid   →  stateDiagram-v2 ...

Validation. Structural constraints — missing fields, ambiguous targets, cross-flow cycles, subflow exit contracts — checked against the specification.

Query. States, transitions, conditions, attributes — ask any question the flow can answer.

Sessions. Init, show, set-state, transition, list. Subflow push/pop for nested workflows. Auto-enters initial subflow on session init. One --session flag turns any command session-aware.

Config. flowr config shows where every value comes from — default, pyproject.toml, or CLI override.

Mermaid export. Generate state diagrams from any flow definition.

Visualise. flowr serve starts a local web server with an interactive D3.js graph — clickable nodes, subflow navigation, zoom and pan. See every state, transition, and guard condition at a glance.

Quick start

Install:

pip install flowr

Requires Python 3.13+.

For the visual editor:

pip install flowr[viz]

Start the server:

flowr serve --path my-project
# Opens http://localhost:8080 — interactive flow browser

Define a flow:

flow: deploy
version: 1.0.0
exits: [deployed, failed]

states:
  - id: prepare
    next:
      ready: execute

  - id: execute
    next:
      success: deployed
      error: failed

  - id: review
    next:
      approve:
        to: deployed
        when: { score: ">=80" }
      reject: failed

Use it:

$ flowr validate deploy.yaml
valid: True

$ flowr states deploy.yaml
prepare
execute
review

$ flowr next deploy.yaml review
state: review
  approve → deployed [blocked]  need: score=>=80
  reject → failed

$ flowr transition deploy.yaml review approve --evidence score=85
from: review
trigger: approve
to: deployed

$ flowr session init deploy-flow
flow: deploy-flow
state: prepare
name: default

$ flowr --session transition approve
from: prepare
trigger: approve
to: review

$ flowr session show
flow: deploy-flow
state: review
name: default
stack: (none)

$ flowr config
project_root = /my/project  (default)
flows_dir = .flowr/flows  (default)
sessions_dir = .flowr/sessions  (default)
default_flow = main-flow  (default)
default_session = default  (default)

Who is this for?

Specification adopters

You build editors, visualizers, CI systems, or orchestration layers. You need a structural format for workflow state machines — not an execution engine. The flowr specification gives you states, transitions, guard conditions, and subflows in a declarative YAML format with a conforming validator. Any tool can parse it. Read the spec.

Agent operators

You run flowr check, then flowr transition, then flowr check again — each time passing the flow name and current state by hand. Or you let sessions track it: flowr session init deploy-flow, flowr --session transition approve, flowr --session check. The session file remembers where you are. Push into a subflow; pop back out. No state to reconstruct, no context to pass.

Developers

You write a flow YAML. You need to know: is it valid? Which states can I reach from here? Does this transition have guard conditions? flowr validate, flowr states, flowr next, flowr check answer these questions — from the terminal for humans, from the Python library for tools.

CLI Reference

<flow> accepts a file path or a short flow name (resolved from .flowr/flows/). Use --flows-dir to override the configured flows directory. All query commands accept --json for machine-readable output (not serve). Evidence: --evidence key=value (repeatable) or --evidence-json '{"key": "value"}'.

Architecture

flowr/
├── domain/           # Core domain — Flow, State, Transition, Session, conditions, validation
│   ├── flow_definition.py
│   ├── loader.py
│   ├── session.py
│   ├── condition.py
│   ├── validation.py
│   └── mermaid.py
├── infrastructure/   # Adapters — config resolution, session persistence
│   ├── config.py
│   └── session_store.py
├── cli/              # Primary adapter — CLI commands, resolution, output formatting
│   ├── resolution.py
│   ├── session_cmd.py
│   ├── serve.py
│   └── output.py
├── server/           # Viz server — FastAPI REST API, flow scanner
│   ├── app.py
│   ├── config.py
│   └── scanner.py
├── static/           # Frontend — D3.js HTML/CSS/JS assets
│   ├── index.html
│   ├── css/
│   └── js/
└── __main__.py       # CLI entrypoint — argparse dispatch

Hexagonal architecture. Domain has no infrastructure dependencies. CLI is the primary adapter. Session store is a secondary adapter behind a Protocol port.

Documentation

• Specification — the flowr format with examples and visual diagrams
• Flow Definition Specification — authoritative YAML format reference
• System Overview — architecture, domain model, module structure
• Product Definition — product boundaries, users, and scope

Development

uv sync --all-extras       # install with dev dependencies
uv run task test            # run tests
uv run task test-fast       # fast tests only
uv run task test-build      # full suite with coverage
uv run task lint            # lint and format
uv run task static-check    # type checking

License

MIT — see LICENSE.