hurl-orchestra

AI first API automation testing.
Test your API from login to checkout. Write each step as a small text file, and hurl-orchestra runs the steps in the right order and shows you which step broke.

_{Play with it on the docs home page: select a note, make it fail, watch the run.}

Say what each step needs, and the order follows

---
id: profile
deps: [auth]
---
GET {{base_url}}/profile
Authorization: Bearer {{auth_token}}
HTTP 200

pip install hurl-orchestra
hurl-orchestra ./tests

I built hurl-orchestra for the moment a Hurl test suite outgrows one file. Each file says which steps must pass first, and the tool works out the rest: it runs the steps in the right order, runs independent steps at the same time and passes values like a login token to the steps that need them. When one step fails, only the steps that depend on it stop, so one run shows every broken flow, not just the first one.

Why teams pick it

• AI first. Plain-text requests, a few frontmatter fields, a JSON plan (--dry-run --json) and CTRF results. A coding agent can write a test, check the graph and read the failure without a human in the loop. There is no model inside and nothing to configure. Use it with your AI assistant →
• Declared, not scripted. deps replace the shell script that calls Hurl in the right order.
• Captures that do not collide. auth captures token, and its dependents get auth_token.
• Honest about flaky failures. known_failures tolerates a failure you understand, with a reason and an expiry date, and the report still shows it.
• Polite under load. A retry policy backs off exponentially, with jitter, and honours Retry-After.
• Ready for CI. A zip of every Hurl report on every run, and a CTRF file for the GitHub job summary.

Under the hood

• Small. Python 3.11+, one dependency (python-frontmatter), and the standard library's graphlib and thread pool for the waves.
• Safe with secrets. Captured values reach Hurl through a private variables file per call, never the process list.
• Fails early. The whole graph is validated before the first request: unknown deps, cycles, wrong types, malformed tolerances and bad regexes.
• Tested. 250+ tests, with ruff and fully typed mypy checks in CI.
• Docs written for people and agents. Short procedures in Simplified Technical English, checked by Vale, and published as llms-full.txt too.

Read the docs

Get started · Concepts · How-to · Reference · Troubleshooting

Say hi

If hurl-orchestra saved your suite from a pile of shell scripts, or you have a flow it cannot express yet, I would like to hear about it. Open an issue, or read CONTRIBUTING.md to send a change.

_{MIT License · Made at Esta couve flor}