Product Updates

Adds scoped in-conversation search, realtime assistant text-input tools, a built-in video prompting skill, and reliability polish for video and media workflows.

Adds fal.ai video generation, ephemeral private cloud media staging, richer video cost tracking, and cleaner media references for generated and uploaded videos.

A small GPT Image 2 follow-up with a minor database migration fix and a few polish updates.

This release adds support for OpenAI's breakthrough GPT Image 2 model, including richer image generation controls, reusable image IDs, and hover previews. It also improves the usage dashboard, TTS playback controls, and a few everyday reliability edges.

This release adds built-in help inside Local Hoist, makes Anthropic tool use and streaming more reliable, improves how shell sandbox setting changes take effect, and fixes auth and cost-reporting issues.

v0.0.29 was a short-lived internal milestone on the way to v0.0.30. See v0.0.30 for the full combined release notes covering everything shipped after v0.0.28.

v0.0.28 is a maintenance release with behind-the-scenes security improvements and a small fix for Anthropic requests.

v0.0.27 adds support for Anthropic's new Claude Opus 4.7, including the new xhigh effort tier and the summarized-thinking opt-in required to keep visible reasoning on by default. It also cleans up Anthropic long-context pricing and context windows to match Anthropic's current documentation, and fixes a small issue with manual image compression runs.

v0.0.26 makes image-heavy chats more resilient, improves a few important tool and sign-in workflows, and adds third-party open source license attribution accessible from the app. This release adds automatic fallback for models that cannot accept image input, improves multi-path file search feedback, and fixes a device-verification sign-in issue on newly recognized devices.

v0.0.25 makes image-heavy chats more resilient and improves a few important tool and sign-in workflows. This release adds automatic fallback for models that cannot accept image input, improves multi-path file search feedback, and fixes a device-verification sign-in issue on newly recognized devices.

v0.0.23 introduces Hooks, reusable workspace automations for running terminal commands from chat. Run them on demand from the composer, attach them to a send, or define conditions so they trigger automatically based on what the assistant did during a turn.

v0.0.22 adds major new chat workflows for skills and file references. This release introduces a workspace Skills manager, slash-skill references in the composer, and @mentions for files and folders with autocomplete, while also improving update messaging and fixing a few cases where chat and tool results could be incomplete or misleading.

v0.0.21 introduces release notes across Local Hoist, so it’s easier to see what changed when you install an update. This release adds a public release history, per-release pages, and in-app release notes surfaces for updates and first launch.

v0.0.20 focuses on making voice capture recovery and packaged desktop behavior more dependable. Recover Audio now does a better job restoring cancelled longer recordings, and packaged builds now handle key runtime state more cleanly.

v0.0.19 gives automatic context compression a clearer, more user-controlled flow. You can now review a pre-compression prompt with options to compress now, defer it for a set number of messages, or turn automatic compression off, and Compression Settings are simpler to understand and manage.

v0.0.18 makes model selection clearer and more dependable, with recommended models surfaced by default, stronger ranking, and cleaner fallback behavior when a previously chosen model is no longer available. It also improves sandbox controls and related product guidance, while updating BYOK pricing copy to better explain current defaults and costs.

v0.0.17 makes Local Hoist easier to use during sign-in, gives desktop users quicker access to key panels, and refreshes the pricing experience with clearer guidance around Managed Credits and BYOK. It also improves device-authorization handling and makes Stop behavior more dependable during active runs.

v0.0.16 delivers a smoother desktop sign-in experience and more dependable stop/start behavior during active runs. Signing in now starts from an email-first flow in the app, device verification stays inside Local Hoist when needed, and active runs recover more cleanly when you stop and continue.

v0.0.15 makes long-running tool workflows more dependable and improves usage reporting accuracy. Web research runs now stop more cleanly, tool results stay in sync more reliably during long-running work, and dashboard token totals better reflect real usage.

v0.0.14 is a focused fix for image-tool workflows. Approved image results now insert their preview message into the conversation more consistently, so generated or opened images appear in chat as expected.

v0.0.13 delivers a cleaner chat layout and more dependable OpenRouter compatibility across model providers, along with additional production-build hardening for published desktop builds. Tool-enabled conversations should behave more consistently, especially when routing through OpenRouter.

v0.0.11 focuses on smoother usage-limit handling, more dependable automatic context compression, and cleaner conversation controls. Rate-limit issues are now surfaced more clearly in the app, and long-running conversations behave more predictably during compression and expand/collapse actions.

v0.0.10 makes Local Hoist more dependable during real coding workflows. This release adds configurable tool sandboxing with stronger shell isolation on macOS, keeps more model settings tied to each conversation, and improves recovery when turns are stopped, retried, or rolled back.

v0.0.7 makes model selection clearer and more dependable, especially when comparing premium or capability-limited models. This update adds better in-context guidance in the model picker, tightens reasoning-effort compatibility handling, and improves built-in file search filters used during coding workflows.

v0.0.6 sharpens Local Hoist's realtime voice experience and makes cost reporting clearer. This update improves realtime setup and defaults, adds better guidance around OpenAI API key requirements and pricing, and makes realtime session summaries and long-conversation cost views more accurate.

v0.0.5 is a performance-focused update that makes long conversations faster to open and steadier to navigate. Opening large chats is smoother, scrolling stays more stable as content settles in, and saved conversation views apply more cleanly during load.

v0.0.4 gives Local Hoist clearer usage visibility and smoother conversation management. This release expands the cost and usage experience, improves browsing for archived and sub-agent conversations, and makes several everyday workflows more dependable.

v0.0.2 makes background work easier to track and conversations more dependable. Background conversations now show clearer live status and completion indicators, and send and tool-run handling is more resilient.

Our first public release established the core Local Hoist desktop experience with automatic updates, better conversation performance, stronger model and cost controls, and a more polished chat workflow. It also introduced important reliability and safety improvements that made the app feel much more production-ready out of the gate.