v0.7.14

- fetch matching `gguf-py` from the region-appropriate llama.cpp source mirror when local gguf is stale or missing
- align the bundled llama.cpp converter metadata with upstream tag `b8724` so release freshness checks pass against the current supported converter ref

v0.7.13

- fetch matching `gguf-py` from the llama.cpp source mirror when the system gguf package is stale or missing
- use Gitee in CN regions and GitHub elsewhere, then retry SafeTensors to GGUF conversion once automatically

v0.7.12

- auto-upgrade stale `gguf` or `transformers` installs during SafeTensors to GGUF conversion and retry once
- align the bundled llama.cpp converter metadata with current upstream support (`b8722`)

v0.7.11

- auto-upgrade stale `gguf` or `transformers` installs during SafeTensors to GGUF conversion
- retry conversion once after the dependency repair step so newer architectures recover without a manual pip fix

v0.7.10

- refresh the bundled llama.cpp SafeTensors-to-GGUF converter to b8721
- block releases that would ship a stale bundled converter by mistake

v0.7.9

- Fix local `Qwen/Qwen3.5-0.8B` so text and image chats default to non-thinking mode under llama.cpp.
- Stop runaway reasoning output that made local replies look stuck or image-blind.
- Recreate the missing temp script directory for AI apps.

v0.7.7

- fix local Qwen/Qwen3.5-0.8B chat so text and image prompts default to non-thinking mode under llama.cpp
- stop runaway reasoning output that made local replies look stuck or image-blind

Release v0.7.5

v0.7.4

v0.7.3

v0.7.2

v0.7.1

- Detect locally installed AI apps and avoid conflicting reinstalls from AI Apps.
- Backfill legacy csghub-lite managed installs so update and uninstall keep working after restart.
- Keep OpenClaw launch handling aligned with detected install paths.

Release v0.7.0