This release focuses on the Core: optimization and stability, with a few new things on top. Faster startup and chat switching: Boot to first token time is down 35%, and opening a chat is 4x faster than the previous version. Embedding warm-up now overlaps chat restoration and startup services, and chat rendering proceeds independently of model, context-usage, and goal refreshes. Paradee Heart local speech: A 9MB text-to-speech model, added alongside Kokoro and made the default for unconfigured settings. Existing voice selections are preserved. It is English-only for now, a starting option. Cost controls: Configurable context breakdown and usage rows, plus selectable price currencies with cached exchange rates. Browser configuration: Streamlined quick settings, centralized extension management, Chromium management pages, download preferences, and optional annotation screenshots. Scheduler editing: Resizable and expanded prompt fields, drafts preserved after failed saves, correct initial state and project inheritance, and optional YAML task storage. Subagents and chats: Directly delegated agents show their working state and can be stopped independently. Parent cancellation propagates correctly, and nested children appear in the sidebar. Fixed disappearing chat and task lists when Sidebar Folders is disabled. Manual chat names stay intact under always-on automatic naming, and rejected custom names no longer disable it. Model preset editing reloads current presets before opening, so stale drafts cannot delete presets created elsewhere, and late selector responses cannot overwrite another chat's selection. Models and providers: Prompt caching preserves cacheable prefixes in long Responses conversations, adds cache routing for Venice and the Agent Zero API, and accepts configurable hints for custom providers. Better model compatibility with reasoning-field fallback, Claude prompt adaptation across third-party providers, and corrected repeat detection for native tool calls. Nebius, Venice, Z.AI/Coding, and OpenRouter embeddings now route through LiteLLM instead of custom endpoints.

I have been unable to update Launcher to 1.7. Click to update 1.7, just hangs for hours.

This release focuses on the Core: a redesigned sidebar, a round of optimizations, and many fixes across the board. New sidebar organization: Chats and scheduled tasks now live in collapsible project folders. You can filter, sort, reorder with drag and drop, move items between projects, and pin the folders you use most. Deeply nested workers are shown in the sidebar, and subordinate names plus project and profile model presets are preserved when delegating or continuing work. Speed and optimizations: The prompt cache is more stable across turns, and each model call carries less overhead thanks to faster preset and plugin lookups, fewer unnecessary processing steps, and reused Codex connections. Settings save faster while preserving complete API-key pools, escaped and multiline secrets, and masked values across key-name casing changes. Large source files open much faster in the Editor. Files: Breadcrumb and text path modes, folder autocomplete, a configurable starting folder, clickable rows, and inline creation and renaming. Open browsers are reused and picker state is preserved. The UX of the file browser has improved substantially after the recent redesign. Models and providers: Native model commentary is displayed as streamed thoughts and preserved during Responses history replay. Provider refusals are surfaced clearly, and blocked requests are no longer retried automatically. Saved credentials for local model providers are no longer overridden (this was a huge blocker for local model enthusiasts). Integrations: Telegram and WhatsApp slash commands expanded with command menus, permissions, plugin settings, confirmed compaction, transcript exports, and chat selection. Project- and agent-profile-scoped configuration for Telegram, WhatsApp, and Email, with simpler integration settings. Browser, UI, and reliability: Better mobile composer sizing, terminal readability, touch behavior, and settings navigation, with chat controls accessible through overflow menus.

File browser rebuilt: The File browser now has directory trees, list and icon views, sorting preferences, and its own Settings category. The Editor unifies text and code editing with syntax highlighting, line numbers, search, Markdown previews, and protection against conflicting saves. Remote file access: Browse and edit folders exposed by connected A0 CLI and Launcher instances directly from the Web UI, with granular permission checks. Transfers are hardened with streaming, integrity checks, atomic writes, configurable limits, and safe archive handling. Beta, feedback wanted - A0 Tag: With an Agent Zero instance open in a Launcher tab, press Ctrl/Cmd + Shift + Enter and ask A0 to do anything on your computer from the app you are in. It can also type text directly in place, currently in text editors, email clients, and custom fields. Credential fields are intentionally not supported for security reasons. Tell us what works and what does not. Browser and Desktop: Safari host-browser setup on macOS, Browser tabs that no longer get stuck loading, faster Browser and Desktop startup, and a draggable canvas rail that remembers its position with per-panel desktop and mobile visibility. Security: Fixed WhatsApp media path traversal, Telegram webhook authentication bypass, and IMAP sender-whitelist parsing bypass. If you use these integrations, please update. 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

From hours of troubleshooting, I used 3 different configs to test. "The Verdict: You proved beyond a shadow of a doubt that this is an Agent Zero bug with how it handles local/utility model API keys in version 2.12. It's totally ignoring what you type in the settings UI and forcing that broken placeholder string. Since the UI won't let you clear its localStorage through devtools, your two options are: 1. The Nuke: Quit Agent Zero entirely, delete or move ~/Library/Application Support/a0-launcher to force the app to rebuild its cache from scratch, and type the settings in fresh. 2. 3. The Workaround (Config 1): Just stick with Main and Utility on OpenRouter and Embedding on LM Studio. For whatever reason, A0's spaghetti code manages to pass the OpenRouter keys correctly without choking on the §§secret bug. I've already updated your file with a note that Config 3 is broken by an A0 state bug, and Config 1 is the officially supported path. You successfully chased the ghost out of the machine. It's a bug in their app, not your homelab. I'm glad to help the Agent Zero group. They only need to reach out to me....