Changelog

What's Changed

• Add LFM3 media capabilities and consume the family runtime (#3230) by @jjang-ai
• Let bundle agents stay out of the Orchestrator's delegation list (#3227) by @tpae
• deps: vmlx-swift 085c1c57 — DeepSeek-V4.1 F32 control parameters for every bundle (#3214) by @jjang-ai
• deps: pin vmlx-swift 1410c6f4 (DeepSeek-V4.1 JANGHT keeps F32 control parameters) (#3213) by @jjang-ai

🐛 Bug Fixes

• Save generation defaults per custom agent (#3233) by @jjang-ai
• Remove the Next turn control from the composer (#3232) by @jjang-ai
• Turn on reasoning for Anthropic, Claude Code, and OpenAI models from their live contracts (#3229) by @tpae
• Coalesce chat preview composition and memoize agent routing JSON (#3198) by @sentry[bot]
• Avoid main thread hang parsing router activity timestamps in Credits (#3177) by @sentry[bot]
• Fix AppHang during system appearance changes in ThemeManager.isDarkMode (#3183) by @sentry[bot]
• Move phone chat handoff system readings off main thread (#3221) by @sentry[bot]
• Offload isGitRepository filesystem probe from main actor in KnowledgeManager (#3222) by @sentry[bot]
• Use Foundation character-set search for newline checks in agent descriptions (#3197) by @sentry[bot]
• Preserve template variables and per-turn generation controls (#3226) by @jjang-ai
• Preserve saved settings and external model availability (#3228) by @jjang-ai
• Reflow source-wrapped paragraphs in model cards (#3219) by @jjang-ai
• Keep message content order through image preprocessing (#3224) by @jjang-ai
• DeepSeek-V4.1 alignment: dSpark solo lease, full effort ladder, model_type detection, cached-token usage (#3216) by @jjang-ai
• Track preparation failures and connect resident cache fixes (#3218) by @jjang-ai
• Unify installed model inventory and correct capacity estimates (#3217) by @jjang-ai
• Own allocator cleanup per request and retain failed MTP diagnostics (#3215) by @jjang-ai

🧰 Maintenance

• Pin vmlx-swift c6fee283: DeepSeek-V4.1 decode attention, dSpark governor fix, SSD restore past 16K tokens (#3223) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.27...0.25.28

Downloads

• Osaurus-0.25.28.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.28-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 cb23748f86bf77ff156e9680e6de70c7b58fed8beac5456a67fa7a3349b09fb2

What's Changed

• deps: pin vmlx-swift 00c369f6 (DeepSeek-V4.1 JANGHT bundles, sampled dSpark) (#3211) by @jjang-ai
• DeepSeek-V4.1-Flash: dSpark, thinking toggle, cached-token usage, image input (#3206) by @jjang-ai
• Use typed Responses and isolate helper reports from user memory (#3203) by @jjang-ai
• Surface initial chat-history open failures in storage recovery (#3202) by @jjang-ai
• Preserve restored chat model during catalog discovery (#3200) by @jjang-ai
• Consume native verification router parity repair (#3194) by @jjang-ai
• Release SQLCipher connections after setup failures (#3195) by @jjang-ai
• Consume cache-index ownership and base-M5 dispatch repairs (#3192) by @jjang-ai
• Observe handled Metal interactivity failures with safe runtime context (#3189) by @jjang-ai
• Keep chat pickers attached and consume nonfatal runtime diagnostics (#3185) by @jjang-ai
• Add native System One decision routing for D3 bundles (#3182) by @jjang-ai

🐛 Bug Fixes

• Report cached prompt tokens in the streamed chat usage chunk (#3212) by @jjang-ai
• Fix memory recall and delayed tool output (#3210) by @jjang-ai
• let long-running mcp calls report progress past 120s and poll without loop nudges (#3193) by @RaajeevChandran
• fix queued channel messages dropped when the post-turn drain is denied (#3208) by @RaajeevChandran
• fix iphone rejecting images for remote provider models (#3207) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.26...0.25.27

Downloads

• Osaurus-0.25.27.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.27-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 033e04c816f85c620868f40f074a76409da63428f14dfe081d5ded58f270a315

What's Changed

• Raise release build timeout to 180 minutes (#3181) by @tpae
• Add opt-in local access key for sensitive routes, settable by bundle import (#3178) by @tpae
• Harness runtimes in the model picker, config documents, and bundles (#3180) by @tpae
• Add pluggable harness runtimes: Claude Code and Pi (#3175) by @tpae
• Add osaurus import for Osaurus bundles (#3174) by @tpae
• Add agent templates (file export/import) (#3172) by @tpae

🐛 Bug Fixes

• Fix: Prevent main thread hang in ExternalModelsSettingsView refreshCounts (#2866) by @sentry[bot]
• Fix: Move keychain OAuth token deletion off main thread to prevent app hang (#2925) by @sentry[bot]
• Prevent UI hang on link clicks (#2946) by @sentry[bot]
• FloatingInputCard: stop querying AVCaptureDevice synchronously on appear to fix hang (#3092) by @sentry[bot]
• Seatbelt: cache isSupported check to prevent main-thread hangs (APPLE-MACOS-37V) (#3094) by @sentry[bot]
• Use cached sandbox availability in effectiveAutonomousExec to fix main-thread hang (#3105) by @sentry[bot]
• Chat sidebar: snapshot orphaned workspace pairings once per render (#3111) by @sentry[bot]
• Memoize transcript context token totals across ChatSession renders (#3112) by @sentry[bot]
• Cache HF timestamp formatters and move org model mapping off main actor (#3120) by @sentry[bot]
• Avoid repeated @Published agents lookups in liveAgentSources (#3121) by @sentry[bot]
• fix main thread hang when creating windows by removing pre-layout pass (#3130) by @sentry[bot]
• Remove synchronous ensureLayout calls from terminal display views (#3131) by @sentry[bot]
• Use LCached for ConfigurationView static settings labels to prevent main-thread hangs (#3144) by @sentry[bot]
• Decouple MTP advisory refresh from memory usage ticks to prevent main thread hangs (#3148) by @sentry[bot]
• Pass isDirectory false when resolving custom avatar URL (#3159) by @sentry[bot]
• Cache native channel configs and attachment roots to avoid main-thread hangs (#3160) by @sentry[bot]
• Remove MainActor sort isolation in ChatModelPickerProvider (#3169) by @sentry[bot]
• Let the chat tab strip move the window like the title bar it covers (#3173) by @tpae
• Fix stale agent paths, Browser Use title, dark tab pill and provider glyph (#3171) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.25...0.25.26

Downloads

• Osaurus-0.25.26.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.26-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 a0e75986ed0886cec1a0c06b364237fabf90b504f3becc59f9b61635907d964f

What's Changed

• feat: integrate Flint runtime, owned weights and SSD reuse fixes (#3161) by @jjang-ai
• Remove retired cold-weight tier setting and status APIs (#3156) by @jjang-ai
• Align adaptive drafter loading and remove legacy fixed widths (#3150) by @jjang-ai
• Pin hybrid cache checkpoint RAM improvements (#3147) by @jjang-ai

🐛 Bug Fixes

• Pin lower batch checkpoint memory and shared SSD writer reservations (#3168) by @jjang-ai
• Resolve declared reasoning tiers across bundle metadata formats (#3167) by @jjang-ai
• Honor declared native Off/On reasoning efforts (#3166) by @jjang-ai
• fix: retain durable SSD seed reuse across model reloads (#3165) by @jjang-ai
• Qualify allocator-owned model loading in Osaurus (#3162) by @jjang-ai
• Pin edited-chat SSD resume protection (#3158) by @jjang-ai
• Align RAM gauge with load and delegation capacity accounting (#3157) by @jjang-ai
• Pin prompt snapshot ownership correction (#3155) by @jjang-ai
• Honor explicit SSD cache quotas and show space advisories (#3154) by @jjang-ai
• tie paired phone features to the phone's access key (#3152) by @RaajeevChandran
• Refresh image progress on equal-length status updates (#3151) by @jjang-ai
• Fix image composer pasted prompts, regeneration, and rendering (#3149) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.24...0.25.25

Downloads

• Osaurus-0.25.25.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.25-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 1e75d6eba9ea34dc7d3a5891f00a2e2b2ee75af8666e8ebaa82b0f11effbf5a7

What's Changed

• FLUX 3 Image Klein: generate, edit, references, delegated jobs (#3145) by @jjang-ai
• Pin vmlx-swift #623: Ornith 122B Adaptive MTP by default, colour parity, router tail (#3142) by @jjang-ai
• Ornith 122B: architecture prefill step, pin vmlx-swift#621 cache reuse (#3140) by @jjang-ai
• make phone runs own their mac chat, catch spoken pii and let a phone call set the privacy filter (#3135) by @RaajeevChandran
• Allosaurus 0.2 on 48 GB Macs: pin vmlx-swift adaptive-MTP lookahead, 48 GB test plan (#3129) by @jjang-ai
• Allosaurus 0.2 JANGHT on 48 GB Macs: pin vmlx-swift exact native MTP and decode speedups (#3125) by @jjang-ai

🚀 Features

• add gpt-6.1-sol to the chatgpt model list (#3141) by @RaajeevChandran

🐛 Bug Fixes

• fix slack app messages confirm opening behind the settings sheet (#3146) by @RaajeevChandran
• Never refuse a chat turn from an estimated context overflow (#3143) by @jjang-ai
• Fix blank model search after selecting a lower list entry (#3133) by @jjang-ai
• Correct Qwen hybrid child memory estimates and preserve model residency (#3132) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.23...0.25.24

Downloads

• Osaurus-0.25.24.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.24-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 dfd54a86cc30f4fb2734314b4dced58a42f36933281a3d5682837c2471469934

What's Changed

• Integrate native workflow media controls, video and memory admission (#3104) by @jjang-ai
• Nemotron 3.5 Super-VL and Bolt runtime: vmlx-swift pin and declared reasoning modes (#3119) by @jjang-ai
• ask for a short spoken reply on a phone's voice-call turns, as a note after the conversation that is never saved (#3106) by @RaajeevChandran
• Consume native image batches across API tools and galleries (#3098) by @jjang-ai
• Expose SeedVR2 upscaling and cooperative job cancellation (#3096) by @jjang-ai
• Preserve effective image workflow parameters in app and API results (#3095) by @jjang-ai

🐛 Bug Fixes

• Fix image bundles entering the language-model download path (#3118) by @jjang-ai
• Wire Nemotron Super-VL runtime and bundle media capabilities (#3114) by @jjang-ai
• accept slack messages posted through apps from allowlisted senders (#3113) by @RaajeevChandran
• add reasoning effort control for gemini text models (#3115) by @RaajeevChandran
• fix agent working folder getting cleared when saving the agent editor (#3116) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.22...0.25.23

Downloads

• Osaurus-0.25.23.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.23-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 9df638fdd416e902f12e35e0bbb334ecb5c29d62430907604787db2124886949

What's Changed

🐛 Bug Fixes

• fix(runtime): pin Flash-Next MTP speed restoration (vmlx-swift #593) (#3087) by @jjang-ai
• Count all production compute weights in model and subagent admission (#3085) by @jjang-ai
• Propagate screenshot scrubbing failures before cloud attachment (#3084) by @jjang-ai
• Rank pinned memory text matches by relevance before salience (#3083) by @jjang-ai
• Integrate native Liquid tools and corrected LFM affine/JANGH runtime (#3082) by @jjang-ai
• Chat table: coalesce per-child height notes and stop grapheme-walking block equality (#3080) by @jjang-ai
• Fix Sentry crashes: plugin ABI table (337), settings window constraint loop (1ZM), clipboard poller race (1QM) (#3079) by @jjang-ai
• Speech models: stop a Core ML crash at launch from repeating on every launch (#3076) (#3078) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.21...0.25.22

Downloads

• Osaurus-0.25.22.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.22-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 e8782930a01a70d2560a8d086769eef98db89903dc11c1233db2d531559b40e4

What's Changed

• MiniCPM-V 4.7 + DSpark: pin vmlx-swift, DSpark load warmup, solo lease for drafter requests (#3075) by @jjang-ai
• Pin vmlx-swift #572: background cache writes (shorter Finishing up) and bundled Uno for K2 (#3063) by @jjang-ai
• stop showing injected memory in user bubbles for chats continued from the phone (#3070) by @RaajeevChandran
• fix /api/show 404 for external models listed by repo name (#3069) by @htjworld
• Support Qwen-Image-2.1-Turbo image bundles (#3059) by @jjang-ai
• fix app hangs by decoupling menu items from high-churn app state (#3052) by @sentry[bot]

🐛 Bug Fixes

• PDF body-text edits: don't advise regenerating the PDF with file_write (#3074) by @jjang-ai
• Workbooks: keep formatting and formulas through model edits (#3073) by @jjang-ai
• file_write: remove the 30,000-character content cap and the 64 KiB streamed tool-call abort (#3066) by @jjang-ai
• read nested metadata.context_length so DeepInfra context windows resolve (#3068) by @RaajeevChandran
• Fix app hang from uncached menu localization lookups (#3053) by @sentry[bot]
• make copy button copy only the response (#3067) by @RaajeevChandran
• Fix UI hang from synchronous WhatsApp config loading (#3058) by @sentry[bot]
• fix themes tab hang by moving duplicate fingerprinting off main thread (#3060) by @sentry[bot]
• fix app hang when opening Apple-app URLs from tools (#3065) by @sentry[bot]
• give shell_run the login shell PATH so version-manager toolchains resolve (#3056) by @RaajeevChandran
• fixed plugins tab freezing on scroll (#3057) by @RaajeevChandran
• Fix UI hang from uncached localization lookups in menu getters (#3054) by @sentry[bot]
• fix forced tool_choice 400 on claude 5.5 models and surface openrouter upstream errors (#3055) by @RaajeevChandran
• Correct 0.25.20 appcast notes to match the tagged binary (#3049) by @jjang-ai
• fix file_edit dropping an explicit empty new_string (#3048) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.20...0.25.21

Downloads

• Osaurus-0.25.21.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.21-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 0e9d8b9c1885ea13297e54df947c2fedef48e16914ebb16da40de288e235dbc1

What's Changed

• Add calculate built-in tool for exact math (#3039) by @jjang-ai
• Say why images are unavailable instead of dropping a pasted image (#3021) (#3040) by @jjang-ai
• harden the phone's secure channel and pairing, and expand phone image generation (#3044) by @RaajeevChandran
• Consume merged tool-stream and K2 runtime fixes (#3046) by @jjang-ai
• Clarify judge polarity and ScreenContext rubric while preserving verdict gating (#3033) by @jjang-ai
• Compaction turns reasoning off in a form self-hosted servers read (#3036) (#3042) by @jjang-ai
• move agent action bar back out of the actions tab (#3045) by @RaajeevChandran
• Host MCP servers see version-manager toolchains (#3024) (#3038) by @jjang-ai
• Model RAM estimates exclude the SSD-resident n-gram table (#3043) by @jjang-ai
• Expose bundle-aware speculative defaults before model load (#3029) by @jjang-ai
• Support local grammar-constrained JSON Schema output (#3028) by @jjang-ai
• send the native mtp option to the phone's model picker for bundles with an mtp head (#3020) by @RaajeevChandran
• Integrate K2-Horizon runtime and native reasoning contracts (#3019) by @jjang-ai
• only show a pairing code while the server is running (#3018) by @RaajeevChandran

🚀 Features

• move agent action bar into its own actions tab under general (#3041) by @RaajeevChandran

🐛 Bug Fixes

• fix longer code fences and render labeled prose fences as code (#3034) by @RaajeevChandran
• Pin Flash native MTP verification corrections (#3022) by @jjang-ai
• Pin PromptSection token estimation main thread hang fix (#3011) by @sentry[bot]
• Fix app hang from synchronous UserDefaults write in ChatTabLayoutStore (#3003) by @sentry[bot]

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.19...0.25.20

Downloads

• Osaurus-0.25.20.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.20-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 785ba10b2f14709316e66c20a554484c99bcc78c389159d9c6feb8e72cf93ca0

What's Changed

• Unify chat composer popups with the model picker style and quiet the lone chat tab (#3017) by @tpae
• let the paired phone manage its agents and chats and give it a secure channel on a mac with no custom agents (#2980) by @RaajeevChandran
• Pin qualified GLM required-tool cache restoration (#3002) by @jjang-ai

🚀 Features

• steer channel messages into the running turn instead of waiting for the next one (#3010) by @RaajeevChandran

🐛 Bug Fixes

• Fix non-MLX Sentry main-thread hangs (chat table incremental apply, visibleContent) and Seatbelt NUL validation (#2922) by @tpae
• Show adaptive native MTP controls before model loading (#3006) by @jjang-ai
• queue channel messages that arrive while a turn is running instead of dropping them (#3008) by @RaajeevChandran
• start a fresh chat for every scheduled run (#3007) by @RaajeevChandran
• Pin MTP head-history and lazy-verifier PLE state fixes (#3005) by @jjang-ai
• Pin request-scoped native MTP warmup fix (#3004) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.18...0.25.19

Downloads

• Osaurus-0.25.19.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.19-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 073d22ea80ae194ec4b56d3602b28cd413c714ff8f4bc49dc96095a92a605beb

What's Changed

• Support Qwen-Image-2.1 generation and editing (#2997) by @jjang-ai
• Restyle chat tab strip as a Safari-style track (#2995) by @tpae
• Pin Qwen4Exp JANGH runtime and media cache fixes (#2991) by @jjang-ai
• MCP robustness, elicitation, and regulated-industry connectors (#2990) by @tpae
• Use exact Naive rotary application on qualified M5 Max (#2972) by @jjang-ai

🐛 Bug Fixes

• Pin GLM sparse-indexer device range optimization (#3001) by @jjang-ai
• Pin canonical Qwen stable-prefix cache reconstruction fix (#3000) by @jjang-ai
• Stream Codex one-shot requests so context compaction works on ChatGPT models (#2999) by @tpae
• Render agent avatars at exact pixel size so they stay crisp (#2998) by @tpae
• Serialize image producers and honor completion boundaries (#2996) by @jjang-ai
• Pin native MTP history and sampled verification fixes (#2994) by @jjang-ai
• Pin runtime mixed-gather dtype correction (#2986) by @jjang-ai
• Avoid storage-unlock scheduler waits and stale starts (#2984) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.17...0.25.18

Downloads

• Osaurus-0.25.18.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.18-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 9f521bccfce22df8ce9b8409ae5558d3137eeca609fadba7c2bf575bedb07fd3

What's Changed

• PDF reading-order extraction for flattened forms (file_read pages, page markers) (#2983) by @tpae
• Router announcements feed and control-plane polling cuts (#2982) by @tpae
• International privacy filter presets with region picker (#2979) by @tpae
• Insights: tamper-evident activity log, Tier 1 coverage gaps, and IA redesign (#2964) by @tpae
• Consume Naive attention-mask GPU ranges (#2967) by @jjang-ai
• Seed Osaurus Cloud starter favorites and unify picker Model options column (#2958) by @tpae

🐛 Bug Fixes

• Fix: NativeAssistantActionsView NSButton init causes main thread hang (#2977) by @sentry[bot]
• Preserve requested repeated text without an automatic retry (#2978) by @jjang-ai
• Preserve native PDF glyph and table cell identity (#2976) by @jjang-ai
• Fit resident delegation context to available memory (#2973) by @jjang-ai
• Use corrected mixed affine QMV Metal kernels (#2971) by @jjang-ai
• Consume GLM native history and text cache boundary fix (#2969) by @jjang-ai
• fix code block copy button and chat width shift after copying (#2960) by @RaajeevChandran
• Preserve WhatsApp final responses when the helper exits (#2965) by @jjang-ai
• Pin compiled post-answer cache persistence (#2962) by @jjang-ai
• Keep the chat tab strip inside the chat column on new tab and live sidebar resize (#2963) by @tpae
• Fix Notes -1708 "doesn't understand the count message" in built-in Notes tools (#2961) by @tpae
• Refresh tool grants between turns and recheck dispatch authorization (#2921) by @jjang-ai
• Move Insights snapshots off the main queue and reject stale results (#2926) by @sentry[bot]
• Move response metrics into Inspect response (#2959) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.16...0.25.17

Downloads

• Osaurus-0.25.17.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.17-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 da932900ce980cc305db28c786cce355061ea6f4c0f7c192f78c060b81af4097

What's Changed

• Pin MLX 0.32.3 for the crash-fix release (#2956) by @jjang-ai
• Redesign chat model picker, Cloud browser and Credits menu (#2947) by @losingmyego
• Settings UX/IA cleanup: grouped-form kit, renamed tabs, Tools & MCP Services rework (#2950) by @tpae

🐛 Bug Fixes

• Pin runtime crash containment and allocator synchronization fixes (#2953) by @jjang-ai
• Reject invalid process strings before Foundation launch (#2954) by @jjang-ai
• only show the model switch advisory when leaving a local model (#2952) by @RaajeevChandran
• fix: reject chunk transfers after lane invalidation (#2949) by @jjang-ai
• Surface preparation errors and pin validated logits runtime (#2938) by @jjang-ai

🧰 Maintenance

• improved model banner ux (#2955) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.15...0.25.16

Downloads

• Osaurus-0.25.16.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.16-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 83219444e36ecdc64229bbeb397033384e8c5fca44e9088799b2ac4336a93ca9

What's Changed

• Orchestrator audit remediation: truthful addendum, apply-first config, pool guard, default new-chat agent, compact spawn envelope (#2936) by @tpae
• keep the phone's runs going when its connection drops and let it rejoin or stop them by run id (#2935) by @RaajeevChandran
• Pin the synchronized ErrorBox runtime (#2934) by @jjang-ai
• bring phone chats to the front on return to the mac and send the orchestrator's quick actions to the phone (#2930) by @RaajeevChandran
• send the worked-for span on session turns for the phone (#2931) by @RaajeevChandran
• Update gathered matmul runtime and fix tied-head cache identity (#2917) by @jjang-ai
• Add prompt_working_folder: let the agent open the folder picker when a chat has no working folder (#2918) by @tpae
• Bill the workspace pool by default for own shared agents; show pool chip on local tabs (#2915) by @tpae
• Harden file-edit tools for document drafting (Raptor + grok-4.3 proof) (#2914) by @tpae
• File change history, in-place document edits, and chat window IA (#2907) by @tpae

🚀 Features

• show total response time in the assistant stats footer (#2916) by @RaajeevChandran

🐛 Bug Fixes

• Respect automatic update policy on model notifications (#2940) by @jjang-ai
• Keep decimal settings editable until commit (#2939) by @jjang-ai
• Drain Claude subprocess output after in-flight delivery (#2937) by @jjang-ai
• Recognize weights in symlinked model bundle diagnostics (#2933) by @jjang-ai
• Pin GLM and Naive JANGH runtime support (#2932) by @jjang-ai
• align model updates and osaurus connect cards with the settings card style (#2913) by @RaajeevChandran
• remember chat scroll position per tab (#2911) by @RaajeevChandran
• cap chat minimap height and make the expanded list scrollable (#2912) by @RaajeevChandran
• Keep the chat tab strip clear of the inspector rail (#2910) by @tpae
• fix chat jumping up when a run finishes while scrolled up (#2909) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.14...0.25.15

Downloads

• Osaurus-0.25.15.dmg — standard build. Pick a model during onboarding.
• Osaurus-0.25.15-full.dmg — 3.6 GB, Raptor 0.6 bundled. Chat offline right after install; no model download. sha256 d68f6dfe5f89f84605a08fafa9825e73c06b99754fab5516452e32c2614b4a26

What's Changed

• Clarify background model admission for description backfill (#2901) by @jjang-ai
• added iOS app integrations (#2875) by @RaajeevChandran

🐛 Bug Fixes

• Fix covariant Self default argument in ProductHuntLaunchCampaign init (#2906) by @tpae
• Fix timing export choices for restored chats (#2902) by @jjang-ai
• Preserve generation metrics across chat history reloads (#2900) by @jjang-ai
• fix right-click copy dropping the chat drag selection (#2899) by @RaajeevChandran
• Show generated agent purposes consistently in local pickers (#2898) by @jjang-ai
• Fix description backfill after in-flight prompt edits (#2897) by @jjang-ai
• Pin runtime with composite and recurrent disk-cache fixes (#2896) by @jjang-ai

🧰 Maintenance

• Postpone the Raptor Product Hunt dialogs (#2905) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.13...0.25.14

What's Changed

• Make agent description optional again with background backfill (#2892) by @tpae
• Retire Raptor v0.5 and shorten the What's New notes (#2882) by @tpae
• Make cancellation and runtime-settings tests deterministic (#2885) by @jjang-ai
• Reuse canonical prefill checkpoints and preserve current model discovery (#2884) by @jjang-ai
• Use qualified Raptor q6 decode fusion (#2883) by @jjang-ai
• Reuse completed prefill chunks in batched cache finalization (#2881) by @jjang-ai
• Fuse Spark large-prefill activation through the vmlx runtime pin (#2879) by @jjang-ai
• Pin vMLX batched rotating checkpoint prefix reuse (#2876) by @jjang-ai
• Give core CI tests a separate budget from cold compilation (#2877) by @jjang-ai
• Generate missing agent descriptions from system prompts (#2869) by @jjang-ai
• Require agent descriptions and guide legacy agents through repair (#2867) by @jjang-ai
• Add full DMG build with bundled Raptor 0.6 (light + full releases) (#2868) by @tpae
• Enable MiMo native capabilities and bound resident allocator reuse (#2863) by @jjang-ai

🐛 Bug Fixes

• Report consistent logical lines in file read and search tools (#2894) by @jjang-ai
• Fix integer setting editing and persistence, search, and follow-up parsing (#2893) by @jjang-ai
• Wait on scan barriers without a Task.sleep poll (#2895) by @tpae
• Preserve streamed tool names and align agent execution contracts (#2889) by @jjang-ai
• Preserve native generation defaults and admitted output budgets (#2890) by @jjang-ai
• Check official model updates during menu-only startup (#2888) by @jjang-ai
• Show when legacy models need revision verification (#2878) by @jjang-ai
• Preserve native thinking defaults for agent and tool requests (#2874) by @jjang-ai
• Require saving SSD cache directory changes before clearing (#2873) by @jjang-ai
• Clear the saved custom SSD cache when prefix reuse is disabled (#2872) by @jjang-ai
• Keep agent purpose metadata current across remote and voice consumers (#2870) by @jjang-ai
• Bind inherited worker models to the headless DefaultAgent eval target (#2871) by @jjang-ai
• Preserve app runtime policy in isolated eval runs (#2865) by @jjang-ai
• Fix: Prevent main thread hang when opening MarkdownLinkText URLs (#2864) by @sentry[bot]
• Stop quoting a settings shortcut that is not bound (#2862) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.12...0.25.13

What's Changed

🐛 Bug Fixes

• Fall back from an unavailable or hung Foundation core model (#2861) by @tpae
• Decide Automation permission probes from the error code, not the localized text (#2858) (#2859) by @tpae

🧰 Maintenance

• Move the Raptor Product Hunt launch to Monday (#2860) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.11...0.25.12

What's Changed

• Add two-phase Product Hunt dialogs for the Raptor launch (Sept 2026) (#2855) by @tpae

🐛 Bug Fixes

• resolve canonical MCP tool names within the originating provider (#2857) by @RaajeevChandran
• Preserve null tool calls and prevent primitive JSON serialization crashes (#2854) by @jjang-ai

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.10...0.25.11

What's Changed

• Accept osaurus://open_from_hf?model= as a Hugging Face model deep link (#2851) by @jjang-ai
• Add "Use with Codex CLI" provider card; accept Codex's Responses input shapes (#2850) by @tpae
• Pin vmlx-swift to 028943b7: learned post-answer resume rows with the history boundary kept beside them; oversized snapshots never written (#2846) by @jjang-ai
• Build native Apple app tools into Osaurus with per-agent opt-in (#2835) by @tpae
• Remove the SSD cache capacity popup from the chat (#2841) by @jjang-ai
• Fix SSD cache sizing and current-chat capacity notices (#2836) by @jjang-ai
• Add install-cohort/age retention telemetry (daily_active + app_launched dimensions) (#2833) by @tpae

🚀 Features

• added read only credit balance endpoint to the local api (#2840) by @RaajeevChandran

🐛 Bug Fixes

• Codex CLI: advertise the context window the server keeps; a thread's prompt_cache_key is its session id (#2852) by @jjang-ai
• fill missing properties on mcp tool schemas (#2849) by @RaajeevChandran
• Keep progress visible across the engine tail and label it "Finishing up…" (#2848) by @tpae
• Fix app hang from standardizedFileURL in LocalVisionEvidence (#2844) by @sentry[bot]
• Fix context compaction: chat-model fallback, visible failures, always-available Compact button (#2843) by @tpae
• Keep channel chat tab live across follow-ups; add Focus Chat on Incoming Messages setting (#2838) by @tpae
• fix model picker hover and favorite heart on bottom rows (#2837) by @RaajeevChandran
• Scope output-complete relay to the owning chat step and reset it per model step (#2831) by @tpae
• Fix app hang in ChatSession.sessionRouterCacheStats getter (#2830) by @sentry[bot]

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.25.9...0.25.10