Releases: aws/nova-act
Releases · aws/nova-act
v3.3.316.0
🚀 What's New
- Browser Session Persistence:
BrowserSessionProviderframework for persisting browser session state (cookies, localStorage) across workflow runs, enabling agents to maintain authenticated sessions without requiring manual login on every run
🔨 Improvements
- [CLI] Refactoring & Documentation: Browser CLI refactoring, bug fixes, and documentation improvements
🐛 Bug Fixes
- Tool Error Propagation: Fix tool errors (e.g.,
AgentRedirectError) not being propagated back to the model - Async Tool Call Fix: Fix async
TypeErrorwhen a tool call coroutine throws an exception, preventing serialization of an unawaited coroutine. Fixes #103 - [CLI] CSS Selector Fix: Fix CSS selector regex false positives where bare HTML tag names were incorrectly treated as CSS selectors
- [CLI] Log Capture Fix: Fix
log_captureresource leak ensuringlog.txtis created early in command execution - [CLI] Install Command Fix: Fix
pip installcommand in CLI README to properly quotenova-act[cli] - [CLI] Starlette Compatibility: Fix Starlette 1.0.0 AttributeError in the AgentCore handler template by removing the redundant ping decorator
v3.3.96.0
🚀 What's New
- Browser CLI: Interact with Nova Act directly from the command line — browse, extract data, take screenshots, manage tabs, and control sessions without writing Python code
- Attach to an already-running Chrome instance via
--cdp, or launch your default Chrome with extensions using--use-default-chrome - Get structured JSON output (
--json), minimal output (--quiet), or detailed logs (--verbose) to fit your scripting and debugging needs - Watch browser actions in real time with
--observefor easier debugging and development - After each command, automatically capture an accessibility snapshot and screenshot so you always know the current page state
- Auto-detects your authentication method (API key or AWS credentials) with options to configure via
--auth-mode,--aws-profile, and--region - Browser sessions survive across CLI invocations with automatic recovery and metadata persistence
- Run
act browser doctorto check your environment oract browser setupfor guided first-time configuration - Per-command log capture, per-step accessibility tracking, session recording, disk usage monitoring, and automatic failure screenshots
- Generate Gherkin-based QA test plans from plain English descriptions using
act browser qa-plan
- Attach to an already-running Chrome instance via
v3.3.35.0
🚀 What's New
- Async Support (Preview): Async implementation of Nova Act for seamless integration with async applications and concurrent browser automation
- Trajectory Auto-Dump: Trajectory JSON files are now automatically saved to the session logs directory after each act call, replacing the previous call log format with standardized trajectory data including richer metadata
- Tool Call Results in Logs: Session log JSON and HTML reports now include tool call results for improved debugging visibility
- [CLI] Remote Workflow Visibility: CLI workflow list now fetches and merges remote AWS workflow definitions with local state, showing synced, remote, and local indicators with filter flags
- [CLI] ARM64 AgentCore Support: Docker builds now target ARM64 for AgentCore compatibility when deploying from x86_64 hosts
🔨 Improvements
- Tool Request Interception Bypass: Tools can now temporarily disable the SDK's request interception during execution, preventing browser freezes when external processes need to control the browser directly
- [CLI] Faster Startup: Heavy imports deferred to command execution time for snappier CLI response
- [CLI] Dynamic Dockerfile Region: Replace hardcoded region in Dockerfile with dynamic region resolution so deploys work in any region
- [CLI] Improved Error Messages: Exception chaining preserves full stacktraces for easier debugging
🐛 Bug Fixes
- Path Traversal Fix: Fix safe relative path handling to correctly handle edge cases where the path equals the base directory
- Color Input Events: Color picker inputs now correctly fire input and change events when values are set programmatically
v3.1.263.0
🚀 What's New
- Improved iFrame actuation: Improved handling of cross-origin elements for better dropdown interactions
🐛 Bug Fixes
- SSL stability: Fixed SSL hook deadlock by restricting unroute to interactive mode and improving context cleanup
- Security: Fixed path traversal vulnerability by validating S3 upload paths against base directory
v3.1.157.0
🚀 What's New
- Bedrock AgentCore Browser Sample: Added a new sample demonstrating how to use Nova Act with the Bedrock AgentCore browser.
🐛 Bug Fixes
- String Response Parsing: Fixed issue where parsed string schema responses contained redundant double quotes.
- Dockerfile Improvements: Fixed Playwright installation in the CLI Dockerfile to ensure browsers are properly set up.
🔨 Improvements
- Logging Improvements: Trace logger is now a global singleton, making it easier for downstream applications to redirect and handle logs. Enhanced trace and log output provides more helpful debugging information.
v3.1.89.0
🚀 What's New
- Default Model Change: Updated default model to
nova-act-previewwhen using API key authentication. Learn more in the AWS Documentation
🔨 Improvements
- Dependency Updates: Upgraded
strands-agentsandstrands-agents-toolsdependencies to include bug fixes, performance improvements, and enhanced functionality - Enhanced Error Handling: Added
ActInvalidToolErrorandActInvalidToolSchemaErrorexceptions for clearer diagnostics when tool call issues occur
🐛 Bug Fixes
- Linux Headless Mode: Fixed browser launch failures on Linux systems without graphical displays by automatically enabling headless mode when no display server is detected
v3.1.18.0
⚠️ Important Notice
- Nova Act SDK versions older than 3.0 are no longer supported. Users must upgrade to the latest version to receive security updates and new features.
- The daily API Key quota has been reduced. If you need dedicated quota for a more ambitious project, please migrate your workflow to the Nova Act AWS Service
🚀 What's New
- MCP Tool Integration Documentation: Added README section showing how to integrate tools from MCP (Model Context Protocol) servers using Strands MCP Client
- Browser Dialog Handling Documentation: Added README section showing how to handle browser dialogs (alert, confirm, prompt) with custom Playwright handlers
🔨 Improvements
- Scrolling Behavior: Fix unintended scrolling by checking scroll only in the desired axis
- Clearer Error Messages: Throttle and daily quota errors now provide more helpful messages to guide users
v3.0.157.0
🔨 Improvements
- Enhanced Element Targeting: More reliable element interaction with improved active element detection and consistent deep element location
- Better Display Compatibility: Increased screen resolution tolerance to 20% for improved compatibility across different display configurations
- Error Messaging: Clearer authentication error messages and graceful handling of unsupported viewport dimensions
🐛 Bug Fixes
- Playwright Installation: Fixed installation issues by pinning Playwright version to prevent compatibility problems
- Backend Stability: Resolved string handling and HTTP compression issues in backend operations
- Route Handling: Fix "Route already handled" error by ensuring unroute hook waits for all executing handlers to complete before removing routes, preventing race conditions during page navigation
v3.0.67.0
⚠️ Important Notice
- Support for SDK versions less than 3.0 will end on January 21, 2026
🔐 Security Updates
- SSL certificate validation is now applied to all page navigations
🚀 What's New
- Increased actuation reliability with enhanced focus detection for shadow DOM and iframe elements
🔨 Improvements
- Better authentication misconfiguration error messages with improved guidance and documentation links
- CLI error reporting now preserves full tracebacks for easier troubleshooting
🐛 Bug Fixes
- Fixed CLI deployment issues for AgentCore Runtime connectivity and duplicate workflow execution
v3.0.5.0
- Update pyproject.toml to include the CLI templates in the generated wheel
- Small nit to remove unused # type: ignore on boto3 client