This release is about control and speed: a real stop button, a faster Web UI, and a round of quality-of-life features you'll feel every day. The stop button: Active runs can now be cancelled end to end. There's a stop button in the Web UI, a new /stop API, and a /stop command in the CLI connector. When you say stop, the whole run stops, including everything happening behind the scenes. Speed: The Web UI starts noticeably faster thanks to asset bundling, service-worker caching, extension manifests, compression, and version-aware cache handling. Long chats stay smooth too: messages render lazily with pagination, and message collapsing plus replayed warning grouping were fixed. Quality of life: Chat renaming is now a builtin feature, you can rename them via the chat list, with /rename, and sidebar pinning is also built in. The Files browser supports drag-and-drop moving for files and folders, including guarded multi-item moves with clear drop feedback. Slash commands are discoverable through skills, and the command picker no longer pops up when you type a trailing slash in a path. Launcher update: You can now drag instance tabs to reorganize them or detach them into separate windows. The Launcher also manages your A0 CLI installation from now on: update the Launcher and both stay current automatically. How to update CLI: open a terminal and type 'a0 update' Web UI: open Settings -> Check for updates -> Review update Launcher: close and reopen the app, then click the Update button New here? Install guide: https://www.agent-zero.ai/p/docs/installation Full changelogs: github.com/agent0ai/agent-zero

A focused stability release. If you ever saw your agent freeze mid-reasoning or stop halfway through a response, this one is for you. New provider: Cerebras Cerebras is now a native provider with onboarding guidance, default model selection, and full UI coverage. If you want extremely fast inference behind your agents, setup takes a few clicks. Linux Computer Use hardening: Window targeting now uses stable AT-SPI window IDs with window-scoped snapshots. Focus is verified before typing, and typing fails closed instead of landing in the wrong window. There is also an unambiguous XWayland wmctrl fallback, plus better evidence handling and retry behavior for failed actions on remote targets. Also in this release: Project-scoped HTTP/SOCKS proxy support for the internal Docker browser, with auth and bypass options and runtime restart handling for active sessions. Codex OAuth reasoning_effort forwarding was normalized so upstream providers never receive unsupported field shapes. How to update CLI: open a terminal and type 'a0 update' Web UI: open Settings -> Check for updates -> Review update New here? Install guide: https://www.agent-zero.ai/p/docs/installation Full changelogs: github.com/agent0ai/agent-zero

This release finishes what the Host Access Gateway started: desktop automation with zero terminal work. Computer Use on macOS, fully guided: The Launcher now completes the whole macOS Computer Use setup for you. Permissions are staged (Accessibility, then Screen Recording), you decide per Instance, and the flow recovers cleanly if the app restarts mid-setup. Once permissions are granted, your original desktop request resumes automatically. No terminal, no manual permission hunting. Smarter chat compaction: Compaction of long chats is now resumable and secret-safe. Authorization, evidence, loaded skills, and secret references are preserved even if compaction gets interrupted. Secret redaction is also scoped to credentials only, so normal settings stop getting masked by mistake. CLI improvements: Long chat history is now paged with cached rows, so big chats render fast and older pages load on demand. Model and preset editing is stabilized: changes save to the global Default preset and chat overrides clear after saving. Agent Zero extension slash commands now show up in the CLI command palette. The Python Playwright client ships with release installs, with Browser dependency repair separated from profile preparation. Launcher update: Clearer Host access capability actions, recoverable Browser setup, quieter diagnostics, clearer first-install progress, and normal Instance creation now limited to Agent Zero releases and local builds (arbitrary images remain available in the Advanced developer flow). How to update CLI: open a terminal and type 'a0 update' Web UI: open Settings -> Check for updates -> Review update Launcher: close and reopen the app, then click the Update button Did you know that Agent Zero can install and use browser extensions from the Chrome Web Store directly in its in-app browser? Watch the video. New here? Install guide: https://www.agent-zero.ai/p/docs/installation Full changelogs: github.com/agent0ai/agent-zero

Hey everyone! I may not be able to join today’s Community Call from the beginning, but I’ll try to jump in later during the call if I feel better. Hope you all have a great session, and hopefully I’ll see you there! The call is here -> https://www.skool.com/live/DlyvNKHbyWw

What shipped Agent Zero Core and CLI v2.5, together with the new version of the Launcher, introduce the Host Access Gateway: a bridge between the sandboxed Docker environment and your real machine, embedded directly in the Launcher. What problem it solves Agent Zero runs containerized for safety. Until now, letting the agent access your host system required installing the A0 CLI, opening a terminal session, and authorizing each capability manually (/computer-use for the desktop, /browser host on for the browser, F4 for code execution). Powerful, but a real barrier for non-terminal users. How to use it Every Web UI session you open in a Launcher tab, or in a detached window through the Launcher, comes with a bundled gateway. From the connection panel (computer icon) you control five separate opt-in permissions: - Files read - Files write - Code execution - Use my Browser - Computer Use Computer Use and Host Browser Use no longer need a terminal at all. The gateway works with your local machine or a remote one, and a single Disconnect button revokes everything instantly. Why it's important for Agent Zero The agent stays sandboxed by default. Host access becomes an explicit, granular, reversible decision instead of a terminal ritual. Same security model, much lower friction. How to update Launcher: close and reopen the app, then click the Update button that appears on restart. CLI: run a0 update in your terminal. Web UI: Settings > Check for updates > Review update. Links Launcher release: https://github.com/agent0ai/a0-launcher/releases/tag/v1.3 Core release: https://github.com/agent0ai/agent-zero/releases/tag/v2.5 Install guide: https://www.agent-zero.ai/p/docs/installation