Skip to content
Agents tracked: 502 Releases (7d): 762 Agent status: no reported issuesSince last limit reset: Claude 11d · Codex 9d For AI agents: Markdown Updated Nov 22, 2026

Agent Zero changelog: what's new each month

Every stable Agent Zero release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 12 months covered; the current month updates daily.

November 2026 · so far

2 releases: 2.15 → 2.16

November brought significant architectural improvements to Agent Zero, including extraction of bundled tools into plugins, unified memory operations, and new cleanup and scheduling capabilities. The month focused on cleaner interfaces, better resource management, and streamlined tool architecture.

Action needed
  • Memory tools unified: separate memory_load, memory_save, memory_delete, and memory_forget tools replaced by single memory tool with action argument.
  • Tool log headlines no longer include agent-name prefix.

Highlights

  • Bundled tools are now available as standalone plugins with per-action permission controls.
  • Memory operations consolidated into a single tool with action arguments for simpler integration.
  • Automatic cleanup system removes aged caches, temporary files, and unused browser versions.
  • Scheduled tasks now catch up after downtime and plugins can own their own maintenance tasks.
  • Cleaner chat output with improved tool call rendering and removed agent-name prefixes.

New

  • A0Cleaner automatically cleans up aged downloads, installer files, browser profiles, and temporary captures.
  • Scheduled task catch-up: recurring tasks that were missed during downtime now run exactly once on the next tick.
  • Plugin maintenance tasks can execute through the scheduler without creating an agent chat.
  • Time Travel snapshot expiry enabled by default with configurable retention up to 500 snapshots within 30 days.
  • Idle browser resource release suspends Chromium and display servers after one hour of inactivity.
  • Workspace-relative Time Travel exclusions and file-size limits for history storage.
  • Host workspace tree settings allow independent or inherited limits per workspace.
  • Welcome composer guidance for new users without API credentials.

Improved

  • Dedicated badge rendering for goal, document_query, skills, vision, browser, office, and computer_use_remote tool calls with distinct colors.
  • Per-action tool permissions across all 12 bundled multi-action tools through profiles.
  • Centralized boolean argument coercion removes duplicated helpers and strict-validation error paths.
  • Tool log output is cleaner with agent-name prefixes removed from headlines, warnings, and fallback entries.
  • Catalog thumbnails added for all 47 bundled plugins.
  • Output-limit truncation now triggers dedicated retry warnings instead of misformat retries.
  • Plugin system allows scheduled tasks to execute without model invocation.
  • Skipped-file reporting and improved Time Travel retention controls.

Fixed

  • Memory tool consolidation eliminates inconsistencies between separate load, save, delete, and forget operations.
  • Token limit handling improved to surface truncation warnings instead of retry loops.
  • Missed scheduled tasks now coalesce properly after system downtime.

October 2026

1 release: 2.14

October brought significant performance improvements across startup and chat responsiveness, along with streamlined browser controls and better provider integrations. LiteLLM and OpenAI SDK upgrades were included with automatic migration support.

Highlights

  • Startup and chat loading are now ~27% faster through overlapped embedding warm-up and optimized chat restoration.
  • Browser startup is leaner, avoiding unnecessary Chromium launches and rendering overhead.
  • Native LiteLLM routing simplifies configuration for Nebius, Venice chat, Z.AI/Coding, and OpenRouter.
  • Browser quick settings redesigned to show only essential options, with advanced controls moved to configuration.
  • Fixed response prompt caching for long conversations to maintain cacheable content boundaries.

New

  • Native LiteLLM routing for Nebius, Venice chat, Z.AI/Coding, and OpenRouter embeddings providers.
  • Automatic startup migration for existing Docker instances when upgrading LiteLLM and OpenAI SDK.

Improved

  • Startup time reduced by ~27% through overlapped embedding warm-up with chat restoration and startup watchdogs.
  • Chat-switch latency significantly reduced.
  • Browser startup overhead eliminated by avoiding Chromium launch for empty tab listings.
  • Redundant load waits removed in browser rendering.
  • Native viewer display size now reused instead of recalculating.
  • Browser quick settings simplified to show only preset, Host/Container, zoom, and extensions; extension management and Chromium settings moved to config.
  • Response prompt caching now retains masked extras at original boundaries for long conversations.
  • Removed overlapping tooltips from Browser quick settings popover.

Fixed

  • Chats no longer disappear when Sidebar Folders is disabled and another plugin imports the folders store.
  • Response prompt caching now preserves cacheable content in long conversations instead of discarding it.

September 2026

2 releases: 2.12 → 2.13

September brought major improvements to file handling and workspace navigation in Agent Zero. The month focused on making file transfers more reliable, streamlining the editor interface, and significantly improving file browser performance and usability.

Highlights

  • Unified text and code editing in a single editor with syntax highlighting, line numbers, and Markdown preview, removing the legacy file-editor modal.
  • Redesigned file browser with configurable path bar, folder autocomplete, and clickable file rows—eliminating loading overlays.
  • Fixed file permissions and ownership preservation when applying editor patches, preventing files from being reset to root-owned 0600.
  • Reduced median open time for large files from 22.5 seconds to 0.37 seconds by skipping hidden Markdown previews.
  • Implemented atomic file transfers with SHA-256 verification and proper payload handling for large operations.

New

  • Configurable file browser path bar supporting both breadcrumb and text input modes with folder autocomplete.
  • Consolidated create-new dropdown replacing separate New File and New Folder buttons.
  • Inline rename and folder creation in the shared footer input row.
  • File Browser preferences for sorting, view mode, initial tree visibility, and independent editing and transfer limits.
  • Exposed CLI and Launcher folders in Files with authenticated HTTP transfers and atomic publication.
  • Negotiated WebSocket payload ceilings with symmetric checksummed transfer protocol for large operations.

Improved

  • File tree ancestors stay clickable during nested navigation, and starting folder is now configurable and defaults to /a0.
  • Canvas and picker no longer blank live listings when opening file pickers; live browsers navigate in place for path links.
  • Directory trees in Files and Editor now load independently and lazily with unified styling and lighter panel backgrounds.
  • Workspace HTTP file transfers now atomic with streaming uploads into partial files, size verification, and fsync.
  • Remote connector tool payloads bounded with explicit continuation metadata, rejecting oversized inline artifacts before base64 expansion.
  • Text editor now explicitly supports line edits with all three patch forms documented.

Fixed

  • File permissions and ownership are now preserved when applying editor patches.
  • Opening a file picker no longer blanks a live canvas listing.
  • Median open time for large files reduced from 22.5 seconds to 0.37 seconds.
  • Oversized inline artifacts are rejected before base64 expansion in connector payloads.

August 2026

4 releases: 2.8 → 2.11

August brought major improvements to agent editing, browser interaction, and context management. The month focused on making configuration more intuitive, enhancing the internal browser with live interactivity, and giving users better visibility into token usage.

Highlights

  • New Agent Editor UI with Easy and Advanced modes for editing agent identity, prompts, and tool policies without destructive changes.
  • Interactive browser viewport now renders live sessions via Xpra instead of static screenshots, with shared runtime across chats.
  • Context Window Usage Indicator shows real-time token counts and usage breakdown directly in the composer.
  • Stop button and /stop API let you cancel running agent contexts without deleting them.
  • Project-scoped profiles let you switch between Global and project scope when managing agents.

New

  • Stop button in chat UI and /stop API endpoint to cancel running agent contexts
  • UI asset bundler and service worker for delivering versioned asset bundles with service worker pre-population
  • WebUI extension manifest allowing extensions to inject entry files into the bundle
  • Chat-naming and pin-to-top plugins extracted from core with dedicated APIs and UI modals
  • Slash commands exposed through skills tool for scoped command discovery
  • Interactive browser viewport with live Xpra session rendering and CDP screencast fallbacks
  • Context Window Usage Indicator plugin showing token counts, percentages, and free-space reporting
  • Context Doctor tool to automatically repair malformed tool-call JSON and incomplete responses

Improved

  • Lazy chat rendering and pagination in messages.js for better performance with long conversations
  • File browser now supports drag-and-drop for moving files and folders with multi-item support
  • Unified x-icon custom element replaces scattered icon markup across WebUI and plugins
  • Profile lifecycle management with per-scope availability, duplication, and synchronized switching
  • Tools, MCPs, and Skills now have explicit default switches with per-item policy controls
  • Browser tab persistence and restore per-chat and in shared scopes without crash advisories
  • Responses render with safe Markdown and LaTeX throughout live streaming instead of raw-to-formatted swap
  • Annotation dialog now includes DOM hover feedback, multi-page batches, and compact controls

Fixed

  • Browser model preset now correctly activates after first Browser tool call and clears at monologue end
  • Function calls emitted during streaming are recovered when terminal Responses event omits them
  • Browser dependency reconciliation, cache migration, and Chromium installation consolidated under single setup lock
  • Bot detection hardening by running Chromium headful through Patchright on private Xvfb display
  • Expensive full-response parsing now coalesced behind 128-character or 50-millisecond cadence to improve performance

July 2026

6 releases: 2.2 → 2.7

July brought major stability improvements, expanded model provider support, and refined the user interface across onboarding, chat, and workspace controls. The month focused on crash prevention, responsive design, and seamless integration with new AI backends.

Action needed
  • OAuth account connections no longer auto-populate Main and Utility models; explicit selection required.
  • Default main model changed to openai/gpt-5.6-terra via OpenRouter.
  • Project-scoped preset definitions removed; all presets now use global collection.

Highlights

  • Crash-safe chat persistence now protects conversation data from corruption during interrupted saves.
  • Global model presets streamline configuration across main, utility, and embedding models.
  • LaTeX rendering in chat displays mathematical expressions inline and in agent reasoning.
  • New provider support includes NVIDIA NIM, Cerebras, and Gemini CLI for orchestration.
  • Desktop session hardening and Computer Use verification improve reliability for browser and system automation.

New

  • Gemini CLI registered as an orchestrator backend with headless workflow support.
  • NVIDIA NIM provider available for chat and embedding models via LiteLLM.
  • Cerebras native LiteLLM provider with Chat Completions support.
  • A0 Launcher host gateway integration with file permission scoping and Computer Use approval.
  • Responsive interface control visibility settings for mobile and desktop toggles.
  • Multiline terminal command grouping executes shell input as single compound commands.
  • Consolidated goal tool replacing separate create, get, update, and storage modules.
  • Browser proxy support with bypass rules and optional authentication for Docker browser.

Improved

  • Deferred model setup now happens in-thread, preserving pending prompts across page refresh.
  • Unified onboarding provider selection into a single step covering Cloud, Account, and Local.
  • Utility model selection is now explicit and no longer silently synced to main model.
  • Live screencast delivery now follows producer cadence with binary and slim frame negotiation.
  • Browser frames stay proportional during resize using contain sizing.
  • Bundled slash commands now list in their own section; edits create project or global overrides.
  • Full-document editing replaces heading-based Markdown pagination for better navigation.
  • Shell and SSH process exits now treated as definitive command completion with exit code helpers.

Fixed

  • Stale OAuth model provider no longer appears selected when stored model slot points at non-connected provider.
  • Custom Agent Zero URL schemes no longer misrouted to browser handler.
  • Viewer input restored after screencast stream restarts by preserving active canvas.
  • Notifications no longer skipped during state snapshots by capturing payloads with GUIDs.
  • Chrome Web Store extension installs now handle redundant reinstalls and safe browser restart.
  • Response tool routing preserves active goal overrides when using native Responses output.

June 2026

4 releases: 1.19 → 2.1

June brought major architectural upgrades to Agent Zero, including a new Responses API transport system, parallel tool execution, and a complete overhaul of remote connectivity and document handling. The month culminated in version 2.0 with significant UI and plugin improvements.

Action needed
  • Remote Link renamed to Remote Control; existing remote configurations must be updated to use the new tunnel provider selector.
  • OAuth dummy API keys no longer displayed for unconnected providers; connection status now gates key visibility.

Highlights

  • Native Responses API support with automatic fallback lets agents execute turns more efficiently on compatible providers.
  • Parallel tool calls now execute independent operations concurrently as visible child chats, improving agent efficiency.
  • Remote Control replaces Remote Link with a refactored tunnel system supporting Cloudflare, Microsoft Dev Tunnels, Serveo, and Tailscale.
  • Document query index reuse eliminates redundant parsing and embedding within the same chat context.
  • Unified canvas surfaces consolidate Files, Browser, Editor, and Desktop into draggable floating modals with Focus mode.

New

  • LiteParse parser with legacy fallbacks and adaptive OCR heuristics for improved document handling.
  • Folder attachments in chat composer flatten directories into file attachments while preserving display names.
  • Protected API endpoint to list installed plugins and toggle non-protected ones.
  • Workspace selector in Time Travel plugin with new API endpoints for workspace navigation.
  • Skills security scanner in Settings that runs checks on skill zips before import.
  • Chat tab metadata exposure in connector API returning chat number, name, and project information.
  • Progressive skill loading moves long tool instructions into compact prompts via document-query skill.
  • Secure post-login redirection with same-origin validation against redirect attacks.

Improved

  • Browser tool guidance now instructs agents to reuse existing tabs by URL or title before opening new ones.
  • Long chat histories replay in small bounded WebSocket frames instead of single oversized payloads.
  • OAuth model slots can now switch providers with dropdown selection for independent account usage.
  • Page connector history replay preserves cursor semantics throughout bounded frame transmission.
  • Welcome screen redesigned with shared composer, quick actions, discovery cards, OAuth accounts, and system resources panel.
  • MCP server configuration now supports project-scoped settings with global/project merge semantics and richer UI.
  • LiteLLM global kwargs normalized to ensure proper merging of framework defaults, configured globals, and per-call overrides.
  • Skills persist through chat compaction by storing skill IDs as chat-wide metadata.

Fixed

  • OpenAI-compatible providers rejecting Responses API payloads now correctly retry through Chat Completions.
  • MCP image attachments materialized as scoped artifact files returning real file paths instead of only inline data.
  • Responses tool schemas normalized to include explicit properties object for compatibility.
  • Plain Responses completions now correctly promoted to finished response entries in WebUI.
  • Full log snapshot replays treated as authoritative, clearing message DOM and rendering in backend order.
  • Providers stringifying toolcalls array within toolargs now handled correctly through same validation path.

May 2026

8 releases: 1.11 → 1.18

May brought major improvements to browser automation, desktop control, and document handling, along with new provider integrations and a dedicated Markdown editor. The month focused on stability, security, and giving users finer control over how the agent interacts with their environment.

Action needed
  • ODT, ODS, and ODP are now primary document formats; OOXML remains available as explicit choice.
  • Markdown editor UI removed from Office/Desktop surface; Markdown documents now use dedicated Editor.
  • computeruseremote is now the sole host desktop-control path; linux-desktop targets only Docker/Xpra.
  • Screenshots for browser, desktop, and computer-use are ephemeral by default and context-scoped; only explicit user-initiated screenshots remain durable.

Highlights

  • Multi-tab browser support lets the agent work across multiple tabs opened by websites or user actions.
  • New onboarding wizard simplifies setup with guided provider and model selection.
  • Bring Your Own Browser mode routes actions through a local Chrome instance instead of Docker.
  • Visual verification required for desktop actions—agents must confirm screen changes with fresh screenshots.
  • Dedicated Markdown Editor surface provides full-featured editing separate from Office tools.

New

  • Multi-tab awareness with concurrent cross-tab work and auto-registration of user-opened tabs.
  • Modifier-key support for ctrl-click, meta-click, and key chords like Ctrl+A and Ctrl+C.
  • Shadow DOM content reading for inspecting modern component-heavy web pages.
  • Host Browser mode to route browser actions through local Chrome with remote-debugging.
  • Ollama Cloud and Nebius Token Factory added as OpenAI-compatible chat providers.
  • Platform-specific desktop skills for macOS (AX), Windows (UIA), and Linux (AT-SPI/Wayland).
  • Connector skill activation endpoint to activate skills in live chats.
  • Configurable skills cap to control the maximum number of active skills.

Improved

  • Screenshot previews in tool messages render as clickable thumbnails in chat.
  • Extension management with uninstall and Open button for extension UI pages.
  • Desktop environment now starts only when opened instead of eagerly on mount.
  • Live document reload after edits so LibreOffice files refresh after agent changes.
  • Unified Office canvas controls with shared headers, Save, and file dropdown in both views.
  • Persistent Desktop lifecycle with single Xpra iframe across canvas, modal, and keepalive.
  • Browser history saved as static JPEGs for immutable transcript previews.
  • MCP multimodal content properly preserved as images and resources flow into history.

Fixed

  • XSS vulnerability in chat markdown rendering now sanitized to prevent attacks.
  • PTY file descriptor leaks in code execution prevented by proper cleanup.
  • Stale Playwright contexts detected and cleared before reuse.
  • Self-update no longer fails on broken symlinks in /usr.
  • ARM64 Xpra codec gaps handled gracefully without startup warnings.
  • Chat composer max-height reduced to prevent UI overflow.

April 2026

4 releases: 1.7 → 1.10

April brought major security hardening, a complete browser overhaul, and new integrations for messaging and CLI access. The month focused on performance optimization, expanded AI capabilities, and strengthened platform security.

Action needed
  • Obsolete a0small profile removed.
  • Tool-call knowledge reference renamed to generic framework file.

Highlights

  • New native Playwright browser with live WebUI viewer, tab management, and universal canvas replaces legacy browser agent.
  • Security fixes for SSRF and path traversal vulnerabilities in document fetching and file downloads.
  • ChatGPT/Codex OAuth support and new A0 CLI Connector plugin enable seamless account integration and host connectivity.
  • Time Travel workspace history with Git snapshots lets you preview, diff, and revert workspace changes.
  • Memory hardening with SHA-256 verification and redesigned messaging integration controls for Telegram, WhatsApp, and email.

New

  • Built-in Skills Selector plugin for browsing and activating skills directly in chat.
  • A0 CLI Connector plugin with authenticated HTTP/WebSocket access and capability discovery.
  • a0-setup-cli skill guides setup with container guardrails and fallback install paths.
  • Office canvas with native read/edit support for DOCX, XLSX, and PPTX files and chart creation.
  • Agent Profile switcher with context-scoped profiles and guided creation wizard.
  • Project-scoped LLM presets for per-project model assignment.
  • Universal Canvas with dockable panels for Browser and Office surfaces.
  • Lexical trigger-based skill matching restored for relevant-skill recall.

Improved

  • Compact default prompts reduced to under 3,000 tokens while retaining all key guardrails.
  • Tool calls now dispatch on first fully closed JSON object detection, stopping stream early instead of waiting for all tokens.
  • DirtyJson parser gains completed flag for safe early extraction of tool calls.
  • Browser Agent can now use dedicated model preset instead of main model.
  • Messaging integrations (Telegram, WhatsApp, email) now support shared transport-level commands for context management.
  • Redesigned email, Telegram, and WhatsApp settings with step-based setup flows and provider presets.
  • FastMCP and MCP upgraded for improved security and stability.

Fixed

  • Tool argument validation no longer crashes on empty toolargs dict for calls like scheduler listtasks.
  • SSRF vulnerability in document query remote fetching blocked localhost and non-public IP targets.
  • Path traversal vulnerability in downloadworkdirfile now validates resolved paths against runtime directory.
  • Regression test added to verify assembled default prompt stays within token budget and retains guardrails.
  • Stream parsing improved to detect end-of-sequence at closing brace rather than end-of-file.

March 2026

7 releases: 0.9.8.3 → 1.6

March brought major architectural upgrades to Agent Zero, introducing a plugin platform with a hub for app-store-like discovery, a complete WebSocket rewrite for real-time UI sync, and new communication integrations including WhatsApp, Email, and Telegram. The Skills System replaced the legacy Instruments with a portable standard.

Action needed
  • Default chat model changed to Sonnet 4.6.
  • Static release notes files removed from docs/releasenotes/; external tooling referencing RELEASENOTESDIR must be updated.
  • WhatsApp default mode changed from dedicated to self-chat.

Highlights

  • New plugin platform with Plugin Hub for discovery, installation, and management of third-party functionality.
  • Real-time WebSocket infrastructure replaced polling-based synchronization for faster UI updates.
  • Built-in Email and Telegram plugins, plus WhatsApp integration via Baileys bridge.
  • Skills System with SKILL.md standard for structured, portable agent capabilities.
  • Chat compaction plugin to manage long conversations and optimize context window usage.

New

  • Plugin platform with extensibility across prompts, tools, secrets, APIs, model providers, and lifecycle hooks.
  • Plugin Hub with app-store workflow: discovery, installation, configuration, validation, scanning, and README viewing.
  • Email and Telegram communication plugins with attachment handling and group-chat support.
  • WhatsApp integration plugin with self-chat and dedicated bot modes, QR and pairing-code setup.
  • Skills System replacing Instruments with SKILL.md standard and UI support for importing and listing skills.
  • Chat compaction plugin as built-in capability for managing long conversations.
  • Model config presets and per-chat model switching in the chat input area.
  • Onboarding flow for missing API keys with banner in chat composer.

Improved

  • Caching for Anthropic models and context window optimization.
  • Load up to 5 skills at a time.
  • WebSocket architecture rewritten with new WsHandler and WsManager for reduced complexity and better connection tracking.
  • Plugin scans now run in parallel instead of serialized queue.
  • Plugin README rendering sanitized with DOMPurify to prevent unsafe HTML.
  • Plugin README image paths corrected to GitHub blob/raw URLs.
  • New compact a0small prompt set reducing initial token usage from 10k to 2.5k tokens.
  • Agents now self-correct malformed tool calls instead of immediately erroring.

Fixed

  • Retry on critical error counter bug.
  • Non-ASCII filename uploads causing 500 error.
  • WebSocket security issue with dispatchtoallsids snapshot during concurrent disconnects.
  • WebSocket memory leaks from SID retention and unreferenced broadcast tasks.
  • Chat navigation buttons repositioned as floating overlay.
  • Settings save no longer overwrites API keys set via Configure Models dialog.

February 2026

3 releases: 0.9.8 → 0.9.8.2

February brought a major overhaul to Agent Zero's core architecture, replacing the legacy Instruments system with a new Skills System and modernizing the UI with real-time updates. The month focused on structural improvements, better performance, and a significantly redesigned interface.

Action needed
  • Instruments system replaced by Skills System; existing Instruments must be migrated to SKILL.md format.
  • Default chat model changed to Sonnet 4.6.

Highlights

  • Skills System replaces Instruments with a new portable SKILL.md standard for agent capabilities.
  • Real-time WebSocket infrastructure replaces polling for instant UI state synchronization.
  • Complete UI redesign with process groups, execution metrics, and enhanced step details.
  • Default chat model upgraded to Sonnet 4.6.
  • Improved caching and context window optimization for Anthropic models.

New

  • Skills System with SKILL.md standard for structured, portable agent capabilities.
  • Built-in skills with UI support for importing and listing skills.
  • Process groups to visually organize and collapse/expand agent actions.
  • Step detail modals showing key-value pairs and raw JSON display.
  • Execution metrics including timestamps, step count, and execution time with tool-specific badges.
  • Ability to load up to 5 skills at a time.

Improved

  • Caching for Anthropic models.
  • Context window optimization.
  • UI redesigned with better organization and visibility.
  • Collapsible responses with show more/less functionality.
  • Code blocks and tables with copy buttons.
  • Real-time synchronization between agent execution and UI via WebSocket.

Fixed

  • Retry on critical error counter.

November 2025

1 release: 0.9.7

November brought significant enhancements to Agent Zero's project management capabilities and agent configuration flexibility. The release introduces new tools, improved document handling, and better support for custom workflows.

Highlights

  • Projects management system for organizing agent work and configurations
  • Custom instructions support for tailoring agent behavior per project
  • New Welcome screen and Dashboard for better user orientation
  • Multiple document support in the document query tool
  • Project-specific secrets management for enhanced security

New

  • Projects management functionality
  • Custom instructions capability
  • New Wait tool for controlling agent timing
  • Project-specific secrets storage
  • Welcome screen and Dashboard
  • Support for multiple documents in document query tool
  • Subordinate agent configuration override support

Improved

  • Memory, knowledge, and files integration with projects
  • Document query tool now handles multiple documents at once

October 2025

1 release: 0.9.6

October brought refinements to Agent Zero's core systems, with updates to memory management visibility, Python compatibility, and provider support. The month focused on stability improvements and expanding integration options.

Highlights

  • New Memory Management Dashboard provides visibility into agent memory usage and performance
  • Github Copilot now supported as a provider alongside existing LLM options
  • Python dual installation support enables better environment management
  • LiteLLM now retries on temporary errors for more reliable API communication
  • Refreshed login screen improves initial user experience

New

  • Memory Management Dashboard for monitoring agent memory
  • Github Copilot provider support
  • Python dual installation capability

Improved

  • LiteLLM retry logic for temporary errors
  • Login screen redesign
  • Kali framework update
  • Browser Use component update

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Agent Zero: pricing, features and alternatives · All changelogs