flowproof 0.9.0

Deterministic tests for AI agents: record a run once, replay it with zero LLM calls. Assert which tools an agent called, with which arguments, in which order.

flowproof

Run your AI agent once, keep the recording, and assert against it from then on. flowproof captures the run at the model boundary - every request and every tool-call decision - and serves it back on later runs, so replay makes zero LLM calls. You assert which tools were called, with which arguments, in which order, and which were not. The same engine drives web, desktop and Citrix.

Agents author, a deterministic engine executes. A flow is described in YAML with natural-language steps, recorded once against the live app, and replayed deterministically in CI with zero LLM calls. Built agent-native: every operation returns structured results a program can reason over, and agents can drive record/run/get_trace/heal over MCP (pip install flowproof[mcp], run flowproof-mcp).

# calc.flow.yaml
name: Add two numbers
app: calc
steps:
  - Type 5
  - Press plus
  - Type 3
  - Press equals
  - assert: display shows 8

from flowproof import Flow

flow = Flow("calc.flow.yaml")
flow.record()            # performs the flow live, writes the trace
result = flow.run()      # deterministic replay -> RunResult
assert result.passed

Or from the shell: flowproof record calc.flow.yaml, then flowproof run calc.flow.yaml (add --json for the structured report).

The wheel bundles the Rust engine; no separate install. Targets: web (headless Chromium, any OS), Windows desktop via UI Automation, sap (SAP GUI Scripting), vision (pixels-only for Citrix/RDP), and api (no UI — HTTP/SQL assertion flows, any OS).

Status: early, in active development (v0.2). Record→replay, model-grounded authoring, healing with reviewable diffs, suites, run recordings, and the MCP server all work and are tested in CI.

Docs and source: github.com/automators-com/flowproof