Reports & distillation

Read what the user said alongside the evidence they shared. AI distillation adds a structured summary and suggested next steps while keeping the original feedback available.

What a report contains#

One tap on the widget is one submission. A report carries the tester's message plus the context captured at tap time: the page URL, browser and platform, an optional screenshot, and an optional session replay. Screenshots and replays never count as separate submissions.

The distillation step#

When a report is stored, it is queued for distillation. An LLM reads the raw text and produces:

• Title — a one-line summary instead of a wall of text.
• Category — bug, confusing, idea or other.
• Severity — an AI assessment from low to critical.
• Summary — an interpretation of the message and available context.
• Next step — a concrete suggestion for what to do about it.

Distillation runs asynchronously. The raw report is stored first and always readable; the distilled fields land shortly after.

Many reports, one shared issue#

When people report the same problem, Testosaurus retains every original capture and groups confident matches into a shared issue. The inbox shows the shared analysis and its report count. Open the issue to inspect each report's wording, browser context and available attachments. Counts represent reports, not distinct people.

Exact known repeats reuse the current analysis. Differently worded reports can use the project's selected AI model for matching; useful new evidence can refresh the shared summary. Matching and analysis use your configured AI credit and spending controls. A repeat does not automatically trigger another full analysis or coding run.

Uncertain matches stay separate and may show a possible match. Use Merge or Split to correct grouping. Original reports stay intact, and split membership is pinned against automatic regrouping. Active work or a newer edit can require you to refresh before changing membership. Resolved and dismissed issues do not silently absorb a new incident.

Captured facts and interpretation#

Original feedback, console entries and browser context describe what was submitted. AI reproduction steps and suggested causes are hypotheses to investigate. A screenshot or replay appears only when consent and capture made that attachment available.

Which plans distill#

Free includes €1.00 in monthly AI credit and Pro includes €3.00. AI usage meters input and output tokens at the current catalog base-token rates plus 5%. On supported paid plans, optional overage draws from the prepaid balance after the included credit is used. When available credit runs out or the daily AI spend cap is reached, new distillation stops; collected reports remain readable and exportable. Skipped reports are not automatically reprocessed after a top-up. See Pricing & limits for rates and spending controls.

Choose whether each supported resource stops at its included limit or uses paid overage, and set alert thresholds in the dashboard. See Dashboard & data.

From a report to a proposed fix#

A coding role investigates a distilled report, edits files in the paths you allow, runs your configured tests and creates a draft pull request in the connected GitHub repository. Choose a catalog model that supports tool calls, its instructions, base branch, path scope and per-run and monthly budgets in Project settings → AI & routing. Review and install the GitHub Actions workflow, then open a report and choose Propose fix.

The workspace owner controls role settings, starts runs and can cancel investigation or testing. Automatic runs are optional: explicitly select report categories on a coding role to start it after matching reports are distilled. Ordinary analysis routing does not start coding work. Team members can inspect run progress, test results and the resulting PR.

Reports in one shared issue reuse its coding history and automatic role reservation. GitHub and Linear each keep their own issue link for the chosen repository or team. If a tracker request was interrupted, inspect the tracker and attach the verified existing issue before retrying creation.

All configured tests must pass after the final edit before a draft PR is published. The current runner uses Node 22 containers in your GitHub Actions account, with network access for the optional dependency setup step and no network access during tests. It does not merge changes. Review the diff and test evidence in GitHub before merging.

Coding uses prepaid credit, with the selected model's catalog token rates plus the existing 5% markup. A run reserves its maximum budget and returns unused credit when it finishes. Testosaurus adds no execution fee; GitHub Actions usage is charged by GitHub where applicable. Automatic top-ups and monthly credit can keep that balance funded.

The initial runner supports Node 22 projects. A coding attempt checks any authorized refill before reserving credit. If funding fails or credit is still insufficient, resolve the balance and use Propose fix to retry; automatic funding does not bypass the role's monthly spending limit.