Predictable repeat economics
Healthy replay is deterministic and uses no model calls. Model spend is reserved for compilation or repair.
Automation decision guide
Choose repeatable automation for work that repeats.
OpenAdapt compiles a demonstrated browser workflow into governed replay. Healthy runs are deterministic and make no model calls. When the interface changes, the workflow re-resolves from evidence, proposes a reviewable repair, or halts.
Where OpenAdapt fits
Built for consequential work trapped behind a GUI.
The strongest fit is a browser workflow your team repeats, where the business intent stays stable, a wrong action matters, and a direct integration is not practical.
Review change once
A repair updates the reusable workflow instead of asking a model to rediscover the task on every run.
Know when it stopped
Configured identity and effect checks halt ambiguous work and preserve a run report for review.
Control the runtime boundary
Choose local, managed browser, customer-cloud, or on-prem execution for the supported workflow.
Side by side
Choose by the operating model you need.
Scroll horizontally to compare all approaches.
| Decision point | OpenAdapt | Traditional RPA | Computer-use agents | Browser recorders |
|---|---|---|---|---|
| Best fit | Repeated, consequential browser workflows without a practical API | Broad enterprise automation with mature connectors and orchestration | Novel or changing tasks that benefit from reasoning each time | Simple browser workflows and test automation |
| Authoring | Record a task; compile the demonstration | Build selectors, rules, and flowcharts | Describe the goal and configure tools and guardrails | Record steps or write a script or prompt |
| Healthy repeat run | Deterministic replay; no model calls | Deterministic replay; platform licensing applies | A model plans and acts on every run | Fixed or model-driven replay, depending on the tool |
| When the UI changes | Re-resolve from evidence, propose a repair, or halt | Repair selectors and workflow logic | Reason through the changed interface | Repair selectors or let a model re-infer |
| Failure control | Configured identity and effect checks can halt and preserve a report | Depends on configured platform controls | Depends on the agent platform and its guardrails | Depends on the tool and script |
| Execution boundary | Local, managed browser, customer cloud, or on-prem | Customer infrastructure or vendor cloud | Local or cloud, depending on the provider | Browser with local or cloud services |
| Current coverage | Browser Beta; Windows Experimental; RDP and Citrix Research | Mature desktop and browser coverage | Broad screen coverage; provider-specific | Browser only |
Governed failure
Control the failure path.
OpenAdapt separates deterministic re-resolution, AI-assisted repair, human teaching, and unsupported drift. Configured identity and effect checks decide whether a consequential step continues.
Resolve
Find the target from retained evidence.
Review
Inspect a proposed repair before reusing it.
Halt
Preserve a report when verification fails.
Measured proof
Faster repeat runs without per-run model spend.
On MockMed, the reproducible browser workflow bundled with openadapt-flow, compiled replay completed the same checked task with lower median latency and $0 in model cost per run.
7.6× faster
median run: 4.9s compiled vs 37.5s agent
$0 vs $0.27
estimated model cost per run
0 vs ~24
model calls per run
Compiled $0 model cost per run; agent $0.27 per run.
This benchmark measures repeat cost and latency on one task. We saw the same pattern in a public OpenEMR demo cross-check.
Method, raw results, and rerun instructionsOpenEMR cross-checkDrift and repair evidence
Start with the right category.
Use a direct API when a stable one exists. Choose traditional RPA when connector breadth and enterprise orchestration matter most. Choose a computer-use agent for novel or exploratory work. Choose OpenAdapt when the browser workflow repeats and predictable replay, reviewable change, and governed failure matter.
Test one real workflow.
Measure authoring time, run time, intervention rate, and incorrect-success rate on work your team already repeats.