Changelog

Stay up to date with the latest features, improvements, and fixes.

Version 0.80.0 - October 22, 2026

Improvements

  • Enhanced run diagnostics across Autoworklet by recording error origin details directly onto run documents for failed executions.
  • Improved script parsing and remote host execution reliability through AST-based callback resolution.

Fixes

  • Resolved selector parsing edge cases involving escaped quotes, apostrophes, and quoted parentheses in pseudo-class arguments.
  • Fixed selector evaluation to ensure empty quoted queries match nothing and pseudo-class arguments integrate properly with natural language fallbacks.
  • Corrected error classification in Autoworklet sandboxes to clearly distinguish between internal engine failures and script-level errors.
  • Prevented recorder configuration failures by falling back to default values when unexpected JSON types are encountered in recorder.json.
  • Fixed wait backoff behavior to only engage when every verification check observes an unchanged screen state.
  • Resolved an issue in the chat interface where cards failed to display inline after headless writing concluded.

Version 0.79.4 - October 20, 2026

Improvements

  • Improved application launch speed by displaying the editor interface before recognition models finish loading.
  • Enhanced shutdown responsiveness in Autoworklet by allowing the application to close without waiting for background library synchronization queries to complete.

Fixes

  • Resolved an issue in Visual Teacher to ensure step bounding box coordinates are correctly read and processed.
  • Corrected offloaded detection outputs to consistently return detection results as arrays matching local detection behavior.

Version 0.79.3 - October 19, 2026

Improvements

  • Added the --no-self-healing command-line option to allow disabling automated healing routines during Autoworklet execution runs.
  • Integrated support for hosted run-document self-healing configuration profiles.
  • Enhanced Autoworklet desktop GUI responsiveness by offloading room creation and library synchronization from the main interface thread.
  • Optimized element wait polling performance with screen hash caching to eliminate redundant screen captures.
  • Improved workbench scalability by executing explore and validation workflows as background asynchronous jobs.
  • Enhanced headless chat delivery throughput by decoupling message dispatch from the primary execution thread.

Fixes

  • Strengthened Autoworklet dashboard API security by enforcing per-launch authentication keys and preventing cross-origin localhost key exposure.
  • Secured browser extension integration by requiring WebSocket authentication and restricting host publishing to dedicated ports.
  • Enforced HIPAA compliance requirements by restricting local OCR and Ollama endpoints to loopback connections.
  • Resolved an issue where environment variables from local .env files were not consistently loaded by the Autoworklet CLI on startup.
  • Fixed state management to ensure self-healing flags reliably reset and apply across daemon runs.
  • Fixed chat queue handling to ensure image attachments remain correctly associated with their active execution run.
  • Eliminated sensitive authentication token entries from client-side GUI login logs.
  • Resolved an issue to ensure native host browser processes launch cleanly on frozen application bundles.
  • Prevented application hangs on exit by properly terminating socket connections and background library sync tasks prior to window destruction.

Version 0.79.1 - October 18, 2026

Fixes

  • Fixed an issue in the Autoworklet runner where accessing type(e) within a worklet exception handler mistakenly referenced the typing primitive instead of Python's built-in type function.

Version 0.79.0 - October 18, 2026

Improvements

  • Enhanced Autoworklet locator performance by caching candidate embeddings across unchanged frames and prioritizing exact crops.
  • Optimized vision-based matching by selectively embedding overlap pairs only when text matching is inconclusive and training locator vision only on misses.
  • Added a dedicated Rust ONNX runtime helper in Linux containers for MoCo with support for full-FP32 model manifests.

Fixes

  • Resolved an issue in the Autoworklet CLI where agent scripts did not execute from the caller's working directory.
  • Fixed Chrome CDP connection handling to cleanly stop waiting when the browser process exits unexpectedly.
  • Ensured storage download, URL generation, and deletion requests explicitly reject folder paths.
  • Fixed storage upload destination path normalization to properly handle trailing slashes, omitted destinations with query parameters, and empty targets.
  • Corrected storage file listing behavior for root paths and folder direct child filtering.
  • Ensured Linux container stability by retaining in-process ONNX execution where required for MoCo.

Version 0.78.16 - October 17, 2026

Fixes

  • Resolved an issue in Autoworklet Docker containers where data directory pruning failed on Python 3.11.

Version 0.78.15 - October 17, 2026

Improvements

  • Added the ability to immediately stop active replays using the Autoworklet system tray hotkey.
  • Added support for viewing and restoring previously removed replay events in the recorder.
  • Enabled speech synthesis through the Autoworklet proxy with automatic fallbacks when no sound output device is detected.
  • Optimized CLI frame streaming and worker data transfers to process large responses in linear time.
  • Reduced CPU and resource overhead by preventing ONNX Runtime worker spinning and limiting background inference threads in sandboxes.
  • Improved database connection efficiency across services by reusing client instances instead of recreating them per request.

Fixes

  • Prevented conflicting Autoworklet GUI instances and ensured application state is preserved when scheduled-run windows close.
  • Fixed replay interruption handling so replays halt immediately during countdowns, pre-flight checks, and screen settling periods.
  • Corrected event timestamping and click frame alignment to accurately reflect recorded interactions.
  • Enhanced recorder input tracking to reliably separate individual clicks from drag gestures.
  • Fixed chat message and card delivery ordering during headless and scheduled automation runs.
  • Resolved browser launch issues in containerized environments by dynamically adjusting Chrome shared memory and sandbox configurations.
  • Enforced target path containment during archive extractions in CLI commands to prevent path traversal.
  • Improved error reporting for headless runs, disabled service calls, and missing API credentials.

Version 0.78.14 - October 13, 2026

Improvements

  • Added resource management for Autoworklet Docker sandboxes, including configurable CPU and memory caps.
  • Introduced sandbox pruning via 'awl --target docker prune' along with automated sweeping of stale containers across environments.
  • Optimized conditional wait evaluations to halt redundant browser actions, input checks, and model evaluations as soon as a condition is met.
  • Improved macOS daemon execution performance by lifting single-thread pinning restrictions on PyTorch and OpenMP.
  • Added connection-aware idle detection to prevent premature daemon shutdowns during active sessions.
  • Enhanced the Autoworklet Record Viewer with millisecond-level precision for replay and clipping ranges.

Fixes

  • Resolved deadlocks during parallel wait checks on Linux by synchronizing PyAutoGUI X11 calls under a shared lock.
  • Ensured held keyboard and mouse inputs are safely released if a wait check is interrupted mid-sequence.
  • Corrected Autoworklet CLI behavior to exit with code 1 when 'exists' or 'exists-text' queries find no match.
  • Fixed path resolution and retrieval for local and remote HTML and JSON run reports, including on run failures.
  • Prevented incomplete or truncated model downloads from being loaded and added automatic re-downloading for affected files.
  • Fixed Autoworklet Recorder Settings to accurately display active values and only save modified configuration fields.
  • Ensured the recorder development server shuts down cleanly upon quitting from the system tray.
  • Prevented Docker sandbox port collisions and isolated container lifecycle management across distinct AWL_HOME directories.
  • Corrected daemon idle timer tracking so it begins counting when the server starts accepting requests rather than at import.
  • Removed outdated urllib3 version pins to improve package compatibility.

Version 0.78.13 - October 13, 2026

Improvements

  • Enhanced Autoworklet CLI stability by launching the local daemon as an independent process rather than forking the running command.

Fixes

  • Fixed an issue in the Autoworklet browser extension where load-state polling prematurely halted during page navigations with temporarily missing execution contexts.
  • Resolved an issue in the Autoworklet CLI where remote execution reports (--report JSON) failed to copy back to the local host path.
  • Corrected Autoworklet API behavior to ensure wait_for_any and wait_for_all fully await active in-flight checks before returning.

Version 0.78.12 - October 12, 2026

Improvements

  • Enhanced the Autoworklet browser agent to observe snapshot references alongside visible text and execute actions directly by reference.
  • Introduced a read-only agent fallback for retained canary checks to maintain continuity instead of halting runs abruptly.

Fixes

  • Hardened daemon startup in the Autoworklet CLI to verify responsiveness, detect terminated child processes, and handle stale lock files and disk tokens reliably.
  • Resolved model download and integrity issues by preventing partial RF-DETR model files from loading, eliminating download race conditions, and automatically redownloading corrupted files.
  • Guaranteed atomic writes for streaming downloads and updates to avoid partial file saves.
  • Standardized return types and text extraction across browser automation APIs, including script execution and content scraping endpoints.
  • Corrected LLM usage metering to accurately attribute model consumption across healing routines, report generation, and daemon background processes.
  • Improved boot validation to properly fail worklet runs with missing documents or empty and whitespace-only URLs.
  • Updated Autoworklet dashboard dependencies to patched versions for enhanced security.