Tags

v1.4.11

v1.4.11 — fix missing Codex plugin install command in README

v1.4.10

v1.4.10 — Claude Code + Codex plugin manifests

v1.4.9

v1.4.9 — CI status badge

v1.4.8

v1.4.8 — README badges

v1.4.7

v1.4.7 — harden qwenspeak skill docs (consent + exfil warnings)

v1.4.6

v1.4.6 — publish to ClawHub via clawhub-publish.yml directly

v1.4.5

v1.4.5 — ClawHub skill publishing + .agents/skills layout + security-review fixes

v1.4.4

better layer split

v1.4.3

f

v1.4.2

v1.4.1

v1.4.0

Job queue, retention, module split

  - tts.py -> tts/ package (constants, logging, utils, jobs,
  generation, pipeline, commands, cli)
  - Sequential job queue with fcntl.flock single-worker
  enforcement
  - Time-based job retention (1d completed, 1w all)
  - TTS_MAX_QUEUE and TTS_LOG_RETENTION configurable via installer
  - Path sandboxing: all user paths resolved under /work
  - Silent log exit when empty, follow mode waits for content

v1.3.1

v1.3.1 - flash-attn + lockbox v2.1.2

- flash-attn 2.8.3 prebuilt wheel (no build deps needed)
- lazy torch/qwen_tts imports (fix CUDA fork crash with flash-attn)
- pin lockbox v2.1.2 (SSH sessions get Docker env vars natively)
- runtime ENV vars moved after build layers (better cache)
- rename scripts/qwenspeak to qwenspeak.sh for clawhub
- fix cancel to kill worker process immediately
- remove leftover "cancelling" status check

v1.3.0

v1.3.0 - lockbox v2.1.2 env passthrough

- pin lockbox v2.1.2 (SSH sessions now get Docker env vars natively)
- remove /etc/environment hack from tts wrapper
- rename scripts/qwenspeak to qwenspeak.sh for clawhub compatibility
- add version metadata to SKILL.md
- fix cancel description (kills worker process immediately)
- remove leftover "cancelling" status check

v1.2.0

v1.2.0 - GPU support + async jobs

- CUDA inference via PROCESSING_UNIT=cuda env var
- CUDA 12.6 + cuDNN runtime baked into single image
- GPU selection via --gpus flag (all, 0, 0,1, etc.)
- ROCm compose overlay ready for future AMD support
- async job system: submit YAML, get UUID, poll progress
- list-jobs, get-job, get-job-log, cancel-job subcommands
- per-job log files with get-job-log -f follow mode
- --with-gpu test flag for GPU integration tests
- skill reworked under .skills/qwenspeak/ with SSH wrapper
- lockbox v2.1 file ops documented (move, copy, info, exists, hash, etc.)

v1.1.0

v1.1.0 - NVIDIA GPU support

- CUDA inference via PROCESSING_UNIT=cuda env var
- CUDA 12.6 + cuDNN runtime baked into single image
- GPU selection via --gpus flag (all, 0, 0,1, etc.)
- ROCm compose overlay ready for future AMD support
- --with-gpu test flag for GPU integration tests

v1.0.0

fuck yeah