# Changelog of Browser Use AI Task Runner (`produkdigitalali/browser-use-ai-task-runner`) Actor

- **URL**: https://apify.com/produkdigitalali/browser-use-ai-task-runner/changelog.md
- **Full Actor documentation**: https://apify.com/produkdigitalali/browser-use-ai-task-runner.md

## Changelog

### 0.1.7 - stable beta default - 2026-09-12

- Set Google Gemini 3.5 Flash-Lite as the beta default after the two-page acceptance case passed with correct output.
- Gemini 3.8 Flash remains selectable but returned two consecutive provider HTTP 503 responses during release validation.

### 0.1.6 - aligned API defaults - 2026-09-12

- Aligned internal API-call defaults with the public input form: Google provider and the live-tested Gemini model.

### 0.1.5 - committed-result exit status - 2026-09-12

- Preserve a successful platform exit when a truthful terminal result was durably committed before a cleanup exception.
- Classify API-key rejection ahead of generic wrapped provider gateway status.

### 0.1.4 - deterministic process exit - 2026-09-12

- Replaced `asyncio.run` at the container entrypoint with explicit loop ownership and a hard process exit after durable result publication and Apify SDK cleanup.
- Pending library background tasks are cancelled without allowing them to keep a finished run alive until the platform timeout.
- Added entrypoint regression coverage for background-task cancellation and success/failure exit codes.

### Beta listing preparation - 2026-09-12

- Updated public documentation to actual PPE prices and known provider limitations. Default input form uses tested Google/Gemini pair.
- Resumed smoke ODOqLeEuYvIlz1IrG passed after quota break: correct two-page authors and one charged action.

### 0.1.3 - bounded provider recovery - 2026-09-12

- Live evidence identified intermittent Gemini HTTP 503. Retry only transient model failures once, within existing 35-second model and task deadlines; never replay actions or charges.
- Preserve clean terminal errors when both attempts fail. No retry for auth, malformed or truncated responses.
- 52 local tests passed. Live PPE two-page case passed on build 0.1.9; repeat was blocked by Gemini HTTP 429.

### 0.1.2 - Gemini response reliability - 2026-09-12

- Explicit thinking configuration limited to recognized 3.0/3.1 families. Gemini 3.8 endpoint rejected explicit level during live testing, so that unsupported parameter is omitted.
- Gemini output cap 4096 tokens, still one provider attempt and unchanged hard time/step/charge limits.
- Added fixed error codes for truncated, invalid and empty provider responses.
- 49 local tests pass, including actual wrapper request configuration via mocked transport. Live retests pending.

### 0.1.1 - provider diagnostics - 2026-09-12

- Added fixed, secret-free labels for provider quota, authentication, access, unavailable model and timeout errors captured by Browser Use.
- 43 unit tests pass, including six diagnostic canary cases.
- Initial Gemini 2.5 Flash task failed; diagnostics identified model unavailable. Gemini 3.8 Flash passed a live schema-valid heading extraction on build 0.1.5. Full release gates remain incomplete.

### 0.1.0 - private development candidate - 2026-09-12

- Implemented Python/Apify/Browser Use runner with three BYOK provider adapters.
- Added strict input/output validation, exact-host allowlist, public-IP pinning proxy, execution caps and sanitized result envelope.
- Disabled Browser Use code/file/search tools, telemetry, cloud sync, vision and persistent sessions.
- Added application-level billing/finalization replay protection; PPE not yet activated.
- Added local and cloud validation suites. Live-provider benchmark and full paid-release gates remain incomplete.
- Fixed Apify input metadata validation after initial build rejection.
- Fixed progress reporting for aborted/failed tasks and recursive structured-output redaction.
- Kept Actor private; no Store publication.

### Starter guidance - 2026-09-11

Prepared project guidance from Playbook IV. Original starter ZIP remains unchanged.