Changelog

What's Changed

• Estimate image context tokens from resolution, not byte size (#1511) by @jjang-ai
• Fix test-core CI failure by skipping TCC status probes under tests (#1507) by @jjang-ai
• Restore Swift-format parity on current main (#1500) by @mimeding
• updated chinese translations (#1491) by @ftzahao
• Fix Gemma-4 JANG audio: one-time patch adds fp16 multimodal passthrough to configs (#1506) by @jjang-ai
• Repin vmlx 36aebd42: strip tool-call markers when no tools offered (fixes chat-UI envelope leak) (#1505) by @jjang-ai
• Repin vmlx a6472300: Gemma4 drifted tool-call envelope fix (stops leaked <|tool_call|> markup) (#1504) by @jjang-ai
• Repin vmlx 466c0a10: Gemma4 bf16 global-attention fp32 upcast (fixes long-context <pad>) (#1501) by @jjang-ai
• Repin vmlx: DiffusionGemma prefill-progress + MLXLMCommon Swift 5 language-mode fix (#1499) by @jjang-ai
• Build the live cache coordinator from the resolved memory-safety plan (slider now governs the live KV cap) (#1497) by @jjang-ai
• Let the RAM-safety slider govern the default KV/context cap (seed=nil) (#1495) by @jjang-ai
• Repin vMLX 839d9f55: live prefill streaming + tool-name canonicalization + monotonic prefill %, prefix-cache proven (#1494) by @jjang-ai
• Re-pin vmlx-swift to main (53e7c208): Gemma de-scramble, slider KV, warm prefix reuse (#1493) by @jjang-ai
• Re-pin vmlx-swift to main (357c81d4): DiffusionGemma router crash fix (#1489) by @jjang-ai
• Gemma 4 speed/cache + DiffusionGemma — consolidated (vMLX main 7d9a85fe) (#1488) by @jjang-ai
• Gemma 4 audio + GGUF-class speed: repin vMLX 1ab081eb, fix TurboQuant KV regression, decode performance settings (#1485) by @jjang-ai
• Add local MCP probe diagnostics (#1476) by @mimeding
• Add Gemma cache defaults and prefill checkpoint (#1469) by @jjang-ai
• Stabilize CI-sensitive tests (#1480) by @mimeding

🐛 Bug Fixes

• fixed app hangs (#1509) by @RaajeevChandran
• fixed main thread hang from non-converging layout loop in management window (#1510) by @RaajeevChandran
• fixed hang main thread hangs (#1498) by @RaajeevChandran
• fixed missing prefill UI (#1492) by @tpae
• fixed chat renaming in sidebar (#1482) by @RaajeevChandran
• hide embedding model from showing up for chat (#1484) by @tpae
• fixed app hangs (#1473) by @RaajeevChandran
• prevent switching chat windows from interrupting an in-flight chat (#1481) by @RaajeevChandran

🧰 Maintenance

• visual and interaction polish (#1513) by @RaajeevChandran
• updated model lists (#1490) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.19...0.19.20

What's Changed

• Add CSV table workflow previews and export (#1369) by @mimeding
• updated chinese translations (#1470) by @ftzahao
• Expose applied generation diagnostics (#1465) by @jjang-ai

🐛 Bug Fixes

• fixed main thread app hangs (#1472) by @RaajeevChandran

🧰 Maintenance

• minor UI cleanups (#1479) by @tpae
• cascading esc on chat view states (#1468) by @tpae
• improved model picker view UX (#1467) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.18...0.19.19

What's Changed

• updated chinese translations (#1466) by @ftzahao
• Document Gemma checkpoint memory safety contract (#1462) by @jjang-ai
• Apply memory safety plan to model loads (#1461) by @jjang-ai
• disabled whats new modal (#1458) by @RaajeevChandran
• Expose memory safety status in cache stats (#1460) by @jjang-ai
• Pin vMLX memory safety contract (#1456) by @jjang-ai
• Expose native MTP runtime status (#1455) by @jjang-ai
• Make RAM pressure advisory for local model loads (#1454) by @jjang-ai
• Keep local model discovery responsive (#1453) by @jjang-ai
• Guard Gemma4 required tool failures (#1450) by @jjang-ai
• fixed v5 memory migration aborting on a foreign key constraint (#1452) by @RaajeevChandran

🐛 Bug Fixes

• fixed quit and plugin teardown crashes (#1463) by @RaajeevChandran
• fixed main thread disk hangs (#1457) by @RaajeevChandran

🧰 Maintenance

• added frontier evals and improved harness stability (#1464) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.17...0.19.18

What's Changed

• updated chinese translations (#1437) by @ftzahao
• import third party skills (#1439) by @RaajeevChandran
• Fix MiMo JANG telemetry and N2 media preflight (#1443) by @jjang-ai
• Resolve indexed sharded local model bundles (#1441) by @jjang-ai
• Pin vMLX Gemma4 tool recovery (#1440) by @jjang-ai
• Pin vMLX Gemma4 RoPE and MoE fixes (#1429) by @jjang-ai
• added agent slash command (#1436) by @RaajeevChandran

🐛 Bug Fixes

• fix instant OOM error refusing local model loads (#1451) by @RaajeevChandran
• fixed fatal crashes (#1449) by @RaajeevChandran
• fixed memory repair that silently kept failing on affected installs (#1444) by @RaajeevChandran
• Fix QuickTime video data URL materialization (#1446) by @jjang-ai
• fixed main thread hangs in watcher and theme editor (#1435) by @RaajeevChandran

🧰 Maintenance

• p2p network hardening, added e2e encryption over secure channel (#1442) by @tpae
• add first time chat shown and used telemetry events (#1445) by @RaajeevChandran
• harness hardening with tools and evals (#1438) by @tpae
• harness improvements (#1433) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.16...0.19.17

What's Changed

• Add Gemma4 media capability diagnostics (#1427) by @mimeding
• Add tool capability availability diagnostics (#1358) by @mimeding
• Add workbook workflow validation and export service (#1368) by @mimeding
• Pin vMLX MiMo routed JANG fix (#1425) by @jjang-ai
• Pin vMLX LFM runtime hardening (#1421) by @jjang-ai
• Pin Nemotron vMLX runtime fixes (#1411) by @jjang-ai

🐛 Bug Fixes

• fixed main thread hangs (#1431) by @RaajeevChandran
• fixed chat jumping to the top when a streaming response finishes (#1424) by @RaajeevChandran
• fixed memory that never produced new episodes on long-running installs (#1423) by @RaajeevChandran

🧰 Maintenance

• improved system prompts for better instructions following (#1430) by @tpae
• moved remote provider diagnostics to edit (#1428) by @tpae
• improved external model management (reveal in Finder, prune deleted) (#1420) by @RaajeevChandran
• doc updates (#1419) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.15...0.19.16

What's Changed

• updated chinese translations (#1414) by @ftzahao
• reorder providers (#1413) by @RaajeevChandran

🚀 Features

• Added App Intents (Siri/Spotlight/Shortcuts) - "Ask Osaurus" and "Run Osaurus Agent" (#1415) by @tpae

🐛 Bug Fixes

• fixed main thread hangs (#1418) by @RaajeevChandran
• moved plugin config keychain reads off the main thread (#1412) by @RaajeevChandran

🧰 Maintenance

• minor changes to onboarding welcome screen (#1416) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.14...0.19.15

What's Changed

• Pin vMLX runtime proof harness (#1408) by @jjang-ai
• [codex] Add runtime proof validation model (#1395) by @mimeding
• [codex] Add model compatibility diagnostics (#1372) by @mimeding
• [codex] Add PDF/PPTX workflow previews (#1370) by @mimeding
• [codex] Add business document attachment summaries (#1367) by @mimeding
• [codex] Add workspace write previews and history (#1366) by @mimeding
• [codex] Add provider/auth/network diagnostics (#1364) by @mimeding
• Improve Claude plugin marketplace install outcomes (#1363) by @mimeding
• Pin vMLX runtime cache parser fixes (#1405) by @jjang-ai
• Pin vMLX hybrid SSM cache fix (#1403) by @jjang-ai
• Pin vMLX streamed parser fix (#1401) by @jjang-ai
• Pin vMLX runtime diagnostics update (#1399) by @jjang-ai

🐛 Bug Fixes

• fixed app hangs in streaming tables, oauth sign in and remote tunnel connect (#1409) by @RaajeevChandran
• fixed unresolved app hangs (#1404) by @RaajeevChandran
• Make release reruns idempotent and add appcast-only deploy (#1400) by @tpae

🧰 Maintenance

• updated onboarding screens (#1410) by @tpae
• fixed bug with capabilities selection not persisting (#1407) by @tpae
• Capabilities refactor - remove preflight tool selection (#1406) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.13...0.19.14

What's Changed

• Pin vMLX hybrid SSM cache fix (#1398) by @jjang-ai
• Pin vMLX Nemotron auto BF16 runtime (#1397) by @jjang-ai
• Pin vMLX Nemotron decode conv runtime (#1394) by @jjang-ai
• Pin vMLX cold-tier opt-in runtime (#1393) by @jjang-ai
• Fix Gemma4 chat-template schema normalization (#1357) by @mimeding
• Pin vMLX Ultra no-load speed gate (#1392) by @jjang-ai
• updated chinese translations (#1380) by @ftzahao
• Pin vMLX Nemotron Ultra release boundary (#1391) by @jjang-ai
• Pin vMLX Nemotron Ultra loader hardening (#1387) by @jjang-ai
• Harden Nemotron Ultra media and cache wiring (#1386) by @jjang-ai
• Pin vMLX Nemotron graph evidence runtime (#1385) by @jjang-ai
• auto split full urls pasted into provider host fields (#1384) by @RaajeevChandran
• Pin vMLX Nemotron live proof runtime (#1381) by @jjang-ai
• Pin vMLX Nemotron parser alias runtime (#1379) by @jjang-ai
• Pin Nemotron fallback macro runtime (#1378) by @jjang-ai
• Budget routed JANGTQ load footprint (#1376) by @jjang-ai
• Use Osaurus production vMLX load policy (#1375) by @jjang-ai
• Gate local model loads on available memory (#1371) by @jjang-ai
• updated chinese translations (#1373) by @ftzahao

🚀 Features

• added grok oauth (#1382) by @tpae

🐛 Bug Fixes

• moved identity keychain reads off the main thread (#1388) by @RaajeevChandran
• fixed main thread hangs when connecting MCP providers and rebuilding the model picker (#1377) by @RaajeevChandran

🧰 Maintenance

• updated onboarding screens (#1396) by @tpae
• improved onboarding create agent view (#1390) by @tpae
• updated remote provider cards for improved UX (#1389) by @tpae
• fixed hangs (#1374) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.11...0.19.12

What's Changed

• Fix plugin tool-loop default iterations (#1361) by @jjang-ai

🐛 Bug Fixes

• disable generative greetings globally and moved to per agent configuration (#1362) by @tpae

🧰 Maintenance

• expanded telemetry scope and added telemetry doc (#1360) by @tpae
• fixed fatal crashes & freezes (#1359) by @RaajeevChandran
• updated models details view, discover models from hugging face and LM studio (#1355) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.10...0.19.11

What's Changed

• updated chinese translations (#1354) by @ftzahao
• save chart as image (#1352) by @RaajeevChandran
• added insights button to assistant responses (#1350) by @RaajeevChandran
• Pin vMLX ZAYA CCA cache runtime (#1351) by @jjang-ai
• Pin vMLX MiniMax parser runtime (#1349) by @jjang-ai
• Pin Nemotron Ultra vMLX runtime contract (#1348) by @jjang-ai
• Pin Gemma4 boolean schema vMLX runtime (#1347) by @jjang-ai
• Pin Nemotron Ultra vMLX runtime (#1345) by @jjang-ai

🐛 Bug Fixes

• fixed crash with ModelDetailView (#1346) by @tpae

🧰 Maintenance

• fixed app hangs across permissions, model browsing, audio and chat launch (#1353) by @RaajeevChandran
• updated feature flag logic and moved more features behind feature flags (#1344) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.9...0.19.10

What's Changed

• Fix Gemma4 reasoning, VL recognition, and tool schema rendering (#1342) by @jjang-ai
• Pin Gemma4 nullable tool schema runtime (#1341) by @jjang-ai
• Pin Gemma4 tool parser vMLX runtime (#1338) by @jjang-ai
• Fix Gemma4 runtime CI guards (#1333) by @jjang-ai
• [codex] Reject workbook targets in file_write (#1332) by @mimeding
• Fix Gemma4 chat image attachment routing (#1331) by @jjang-ai

🐛 Bug Fixes

• fixed minor issue with chat template (#1343) by @tpae
• fixed main thread freeze (#1337) by @RaajeevChandran
• added guards to prevent sentry from flagging false hangings (#1334) by @tpae

🧰 Maintenance

• remove unused sandbox features (#1340) by @tpae
• remove old migrations and deprecated code (#1339) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.8...0.19.9

What's Changed

• Add Gemma 4 12B unified vMLX runtime pin (#1329) by @jjang-ai
• opt in telemetry (#1320) by @RaajeevChandran

🚀 Features

• added claude plugins catalog (#1327) by @tpae

🐛 Bug Fixes

• voice transcription improvements (#1326) by @RaajeevChandran

🧰 Maintenance

• fixed onboarding issues (#1330) by @tpae
• updated chinese and german translations (#1328) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.7...0.19.8

What's Changed

• Prove LFM tool runtime and cache reuse (#1314) by @jjang-ai

🐛 Bug Fixes

• fixed issue with context budget being out of sync (#1324) by @tpae
• fixed skill renaming (#1321) by @RaajeevChandran
• bug fixes (#1319) by @RaajeevChandran

🧰 Maintenance

• launch stability improvements (#1323) by @tpae
• moved default tools behind a feature gate (#1318) by @tpae
• improve prompt for better guidance (#1317) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.6...0.19.7

What's Changed

🚀 Features

• added MiniMax provider (#1316) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.5...0.19.6

What's Changed

• Consolidate Step 3.7 and LFM runtime proof (#1310) by @jjang-ai
• Prove XLSX file_read preview metadata (#1312) by @mimeding
• [codex] Add AtlasCloud provider preset (#1304) by @mimeding
• Bound raw file_read previews (#1306) by @mimeding
• [codex] Audit high-fidelity read_file coverage (#1305) by @mimeding
• Preserve PPTX slide table structure (#1307) by @mimeding
• [codex] Validate relay public URL before showing connected (#1302) by @mimeding

🚀 Features

• added stateful agent folder ops (#1313) by @tpae

🧰 Maintenance

• minor UI improvements (#1315) by @tpae
• combined sandbox file operations into unified file operations (#1311) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.4...0.19.5

What's Changed

• Add post-1266 live family cache matrix harness (#1268) by @jjang-ai
• [codex] Prove folder artifact tool exposure (#1295) by @mimeding
• [codex] Restore keychain test formatting gate (#1293) by @mimeding
• [codex] Add ChatGPT auth diagnostics (#1294) by @mimeding
• [codex] Restore Swift format gate (#1286) by @mimeding

🚀 Features

• added ability for sandbox to read host workspace (#1298) by @tpae

🐛 Bug Fixes

• fixed missing stop button (#1301) by @tpae

🧰 Maintenance

• minor improvements (#1297) by @RaajeevChandran
• crash fix: reverted breaking change, added failsafe check (#1296) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.2...0.19.3

What's Changed

• chat polish with smoother streaming, stable tool call animations and scroll cache fixes (#1283) by @RaajeevChandran
• [codex] Harden StatsPack schema metadata (#1280) by @mimeding
• [codex] Harden PPTX ingress metadata (#1279) by @mimeding
• [codex] Harden generated XLSX cell output (#1278) by @mimeding
• [codex] Preview XLSX workbooks through file_read (#1277) by @mimeding
• [codex] Harden MCP OAuth metadata redirects (#1276) by @mimeding
• [codex] Contain skill file paths (#1275) by @mimeding
• [codex] Contain Hugging Face download paths (#1274) by @mimeding
• [codex] Bound document parser adapter bridge (#1273) by @mimeding

🐛 Bug Fixes

• fix MCP OAuth discovery for servers that publish path scoped PRM (#1285) by @RaajeevChandran
• fixed scroll to bottom lag & mid stream jumps during streaming (#1272) by @RaajeevChandran

🧰 Maintenance

• update hubspot oauth to use client id and secret (#1284) by @tpae
• remove restrictive instructions on AGENTS.md (#1281) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.1...0.19.2

What's Changed

• feat(model-browser): add Newest / Oldest sort options (#1270) by @therahul-yo
• Fix zero free space on external model volumes (#1267) by @atomtanstudio
• chat improvements (#1259) by @RaajeevChandran
• Stabilize plugin spec root isolation (#1261) by @mimeding

🚀 Features

• Default Agent is now Onboarding Agent (#1271) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.19.0...0.19.1

What's Changed

• added OpenAI compatible /v1/completions endpoint for FIM autocomplete (#1247) by @RaajeevChandran
• fix: preserve exposeToNetwork setting across launches (#1243) by @miguelAngelo1999
• Pin vMLX runtime and harden local model proof gates (#1242) by @jjang-ai
• allow custom base URL for built in remote providers (#1246) by @RaajeevChandran

🚀 Features

• added privacy filter (#1244) by @tpae

🧰 Maintenance

• added whats new privacy filter (#1252) by @tpae
• moved claude plugins to plugins section (#1250) by @tpae
• improved skills page in settings (#1245) by @RaajeevChandran

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.18.38...0.18.39

What's Changed

• added no timeout option for remote providers (#1237) by @RaajeevChandran
• Pin vmlx main runtime and harden server boundaries (#1201) by @jjang-ai
• persist per turn timing and add two page export options chooser (#1230) by @RaajeevChandran
• Harden plugin artifact installs (#1223) by @mimeding
• Harden plugin static file containment (#1222) by @mimeding
• Redact provider secret header logs (#1221) by @mimeding
• Reject empty XLSX workbook output (#1218) by @mimeding
• Redact clipboard diagnostics (#1217) by @mimeding
• improved export chat UI (#1227) by @RaajeevChandran
• Harden XLSX ZIP package parsing (#1216) by @mimeding

🐛 Bug Fixes

• fixed HF import not surfacing models in the models page (#1234) by @RaajeevChandran

🧰 Maintenance

• add chat rename guard (#1238) by @RaajeevChandran
• onboarding screen language changes (#1229) by @tpae

Full Changelog: https://github.com/osaurus-ai/osaurus/compare/0.18.37...0.18.38