Outway documentation

Monitoring

Track website changes using Outway monitors, schedules and capture evidence.

Start with a successful capture

Verify that your target can be collected before scheduling it. Open Monitors to configure recurring collection and inspect run history.

Compare evidence

Compare the previous and current successful captures. Inspect timestamps, content and failure states before interpreting a change. A failed collection is not evidence that the source removed its content.

Keep recurring usage bounded

Recurring work consumes collection capacity and allowance. Choose a suitable interval and review job history. Project-bound AI keys cannot create recurring schedules; use the authorized workspace workflow.

Automate with the API

Monitor and schedule contracts are included in the OpenAPI reference. Use its exact fields and permissions. Keep webhook secrets private and verify delivery according to the endpoint’s contract. Do not assume a configured schedule proves that every run completed.

Review a change with AI

Where the self-hosted model is enabled, explicitly request a semantic review from a successful changed monitor run. Outway uses the run’s recorded baseline, checks before/after quotations against saved text and retains source hashes. Results include uncertainty; quote matching does not verify the model’s interpretation.

This is an on-demand review with explicit model-use consent and bounded usage. It does not silently run a model on every monitor change or replace collection failure states with an AI conclusion.

Automatic evidence reviews

Workspace administrators can save an automatic review policy from monitor history. Choose an instruction, a daily review limit and explicitly allow saved evidence to be sent to the configured self-hosted model. A separate deployment switch must also be active; the app shows when it is disabled.

Only changed successful runs completed after consent and within the last 24 hours are eligible. Global model limits still apply. Pausing stops new admissions; work already admitted may finish. Requests with an uncertain outcome are never automatically repeated. Review history links to the saved, capture-scoped analysis.

Subscribe a webhook to semantic.reviewed to receive ready-review evidence IDs. Verify signatures and deduplicate event IDs. Before-and-after quotes are checked against saved evidence, but model interpretations remain unverified; this is not a calibrated severity detector.