v0.7.0

0.7.0 (2026-08-22)

✨ Features

• ios: reduce Qwen3 thermal load during generation (#43) (6280e34)

v0.6.0

0.6.0 (2026-08-21)

✨ Features

• ios: add turn-scoped inference primitives (#40) (de6b331)
• normalize LLM generation turn outcomes (#39) (8168b02)
• simplify agent chat experience (#42) (5e0e7bd)
• tts: support configurable speech speed (#36) (7946f41)

🐞 Fixes

• embeddings: enforce MLX input limits and correct masks (OSS-31) (#37) (a1d6faf)
• stt: define and validate the direct STT audio contract (OSS-30) (#38) (ce8ec9b)

v0.5.1

0.5.1 (2026-08-06)

🐞 Fixes

• handle Nitro Modules progress events (#32) (226001d)
• harden MLX generation path with staged errors and terminal stream events (#34) (e14d78b)

v0.5.0

What's Changed

• Improve stability with input validation, download manifests, and Expo 55 by @corasan in #19
• Add LLM generation/context config, history management, and upgrade to Expo 55 by @corasan in #20
• Add ChatSession API and upgrade mlx-swift-lm to 3.x by @corasan in #21
• Add Embeddings module for on-device text embeddings by @corasan in #22
• Stream byte-level download progress with parallel fetches by @corasan in #23
• Add CI workflow for lint, typecheck, and tests by @corasan in #24
• Bump default tokenBatchSize from 1 to 4 by @corasan in #25
• fix(ios): move MLXArray copy off audio tap path by @corasan in #26
• Isolate HybridLLM state on MainActor by @corasan in #27
• Extract shared validateLoadOptions helper by @corasan in #28
• Optimize managed history trimming by @corasan in #29
• refactor(ios): unify generation paths via GenerationSink protocol by @corasan in #30
• Remove hardcoded Memory.cacheLimit override in load by @corasan in #31

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

• feat: Add streamWithEvents with thinking state machine by @corasan in #15
• feat: Text-to-Speech support with PocketTTS by @corasan in #16
• feat: Speech-to-Text support with live transcription by @corasan in #18

Full Changelog: v0.3.0...v0.4.0

v0.3.0

0.3.0 (2026-01-21)

Features

• tool calling #14 f05e6dc

v0.2.2

0.2.2 (2025-12-27)

🐞 Fixes

• prevent concurrent model load crash #12 ca7d714

v0.2.1

0.2.1 (2025-12-27)

• feat: get model info #11

v0.2.0

0.2.0 (2025-12-07)

Features

• add additionalContext option #9 16af631
• message history management #10 262c913

v0.1.1

What's Changed

• fix: exported swift files by @corasan in #8

Full Changelog: v0.1.0...v0.1.1