{"id":51544029,"url":"https://github.com/openadaptai/openadapt-flow","last_synced_at":"2026-07-09T16:01:43.723Z","repository":{"id":370205573,"uuid":"1291376938","full_name":"OpenAdaptAI/openadapt-flow","owner":"OpenAdaptAI","description":"Demonstration compiler: record a workflow once, compile to a deterministic vision-anchored script, replay locally, self-heal on drift","archived":false,"fork":false,"pushed_at":"2026-07-08T16:06:31.000Z","size":966,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-07-08T17:04:13.795Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/OpenAdaptAI.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-07-06T18:40:14.000Z","updated_at":"2026-07-08T16:07:52.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/OpenAdaptAI/openadapt-flow","commit_stats":null,"previous_names":["openadaptai/openadapt-flow"],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/OpenAdaptAI/openadapt-flow","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OpenAdaptAI%2Fopenadapt-flow","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OpenAdaptAI%2Fopenadapt-flow/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OpenAdaptAI%2Fopenadapt-flow/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OpenAdaptAI%2Fopenadapt-flow/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/OpenAdaptAI","download_url":"https://codeload.github.com/OpenAdaptAI/openadapt-flow/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/OpenAdaptAI%2Fopenadapt-flow/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35304875,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-09T02:00:07.329Z","response_time":57,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2026-07-09T16:01:42.149Z","updated_at":"2026-07-09T16:01:43.712Z","avatar_url":"https://github.com/OpenAdaptAI.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# openadapt-flow\n\n[![CI](https://github.com/OpenAdaptAI/openadapt-flow/actions/workflows/ci.yml/badge.svg)](https://github.com/OpenAdaptAI/openadapt-flow/actions/workflows/ci.yml)\n[![PyPI](https://img.shields.io/pypi/v/openadapt-flow)](https://pypi.org/project/openadapt-flow/)\n[![Python](https://img.shields.io/pypi/pyversions/openadapt-flow)](https://pypi.org/project/openadapt-flow/)\n[![License: MIT](https://img.shields.io/badge/license-MIT-green)](LICENSE)\n\nRecord a GUI workflow once. Replay it deterministically, locally, for free.\nA model only touches the script to repair it.\n\n![One demonstration, two UIs, same compiled workflow — the right side self-heals under a theme it has never seen](docs/showcase/demo.gif)\n\n*Real screenshots from the two runs in [`docs/showcase/`](docs/showcase).\nLeft: the UI the demo was recorded on. Right: a theme it had never seen — each\nstep re-resolves through OCR or geometry, and each fix is written back to the\nscript as a reviewable diff. Zero model calls on either side.*\n\n## Try it\n\n```bash\npip install openadapt-flow \u0026\u0026 playwright install chromium\n\nopenadapt-flow demo-record --out rec                     # record a demonstration\nopenadapt-flow compile rec --out bundle --name my-task   # compile it\nopenadapt-flow replay bundle                             # replay: local, $0\nopenadapt-flow replay bundle --drift theme               # drift the UI, watch it heal\n```\n\nThe last two commands serve the bundled MockMed demo app and write an\nillustrated `REPORT.md` per run. Pass `--url` to replay against your own app;\nrecorded parameter values are the defaults and `--param` overrides them.\n\n## How it works\n\nComputer-use agents re-reason through your task with a large model on every\nrun. That's the right shape for a task nobody has automated before, and the\nwrong one for the 500th referral this month. openadapt-flow compiles the\ndemonstration instead.\n\nEach compiled step carries a template crop, an OCR label, geometry landmarks,\nand postconditions derived from what the demo actually changed on screen. At\nreplay time a resolution ladder tries them in order: local template match,\nglobal template match, OCR, landmark geometry, then (optionally) a grounding\nmodel. Healthy scripts never leave the first rung. Milliseconds, no model\ncalls, no per-run cost.\n\nWhen the UI drifts, a lower rung still finds the target and the fix lands in\nthe bundle as a diff you can review. When the screen stops matching\nexpectations entirely, the run halts with a report instead of guessing, and\nsteps tagged irreversible won't act on a low-confidence match at all.\n\nThe runtime is vision-only (PNG in, clicks and keys out) behind a small\n`Backend` protocol. The reference backend is a headless browser, which is why\nthe whole loop runs in CI with no OS permissions. Desktop and RDP backends\nare adapters to come, not rewrites.\n\n## Proof\n\nEvery CI run records a demonstration, compiles it, and checks:\n\n| Scenario | Outcome |\n|---|---|\n| Baseline replay ×3 | all steps `template` rung, 0 heals, 0 model calls |\n| Theme drift | succeeds; 8/8 anchors healed; healed bundle replays clean |\n| Moved buttons | succeeds via global template search |\n| Renamed buttons | succeeds via landmark geometry |\n| Surprise modal | fails loudly, naming the violated postcondition |\n| Non-recorded parameter | substituted and verified by OCR of the final screen |\n\nArtifacts: [baseline run report](docs/showcase/baseline-run/REPORT.md) ·\n[theme-drift run report](docs/showcase/theme-drift-run/REPORT.md).\n\nThe same loop has also run against a real third-party app: the official\nOpenEMR public demo (fake patients only, resets daily). An 18-step clinical\nworkflow — log in, find a patient, scroll a dense dashboard, add a\nparameterized note — replayed **5/5 in fresh browsers with zero model\ncalls**, scrolling closed-loop (each SCROLL step scrolls until the next\nanchor actually resolves, so content growth between runs can't displace the\ntargets below it). Full runs, failure analysis, and honest caveats:\n[docs/showcase-openemr/FINDINGS.md](docs/showcase-openemr/FINDINGS.md).\n\nCompiled workflows can also be emitted as Agent Skills or MCP servers\n(`emit-skill` / `emit-mcp`), so other agents can invoke them.\n\n## Benchmark\n\n![Latency and cost: compiled replay vs computer-use agent](benchmark/latency_cost.png)\n\nWe ran the same MockMed task both ways on 2026-07-08 with the same OCR\nsuccess check: 100 compiled replays against 20 runs of a claude-sonnet-5\ncomputer-use agent. Both arms went 100 for 100 and 20 for 20, so on an app\nthis simple the story isn't success rate. It's that a compiled replay\nfinishes in 4.9s (p50; 5.1s p95) with zero model calls, while the agent\ntakes 37.5s (p50; 43.4s p95) at about $0.27 per run at list price, every\nrun, forever. Full numbers, methodology, and caveats:\n[benchmark/BENCHMARK.md](benchmark/BENCHMARK.md).\n\n## Status\n\nv0: 163 tests, drift matrix in CI. Solid for the reference browser backend.\n`DESIGN.md` has the module contracts; `docs/L1_INTEGRATION.md` covers feeding\nlayered clinical-data platforms.\n\n## Development\n\n```bash\ngit clone https://github.com/OpenAdaptAI/openadapt-flow \u0026\u0026 cd openadapt-flow\npip install -e '.[dev]' \u0026\u0026 playwright install chromium\npytest -q\n```\n\nThe demo GIF is generated from real run artifacts by\n`scripts/make_demo_gif.py`. MIT license.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fopenadaptai%2Fopenadapt-flow","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fopenadaptai%2Fopenadapt-flow","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fopenadaptai%2Fopenadapt-flow/lists"}