Skip to content

v0.4.5

Choose a tag to compare

@github-actions github-actions released this 18 Mar 02:54
· 409 commits to main since this release
Bugs fixed:
- Fix infinite loop when Gemini returns only Thinking blocks (#704)
- Fix tool_blocklist not detected on daemon restart (#666)
- Fix MCP servers not receiving credentials from .env/vault (#660)
- Fix image base64 causing compaction storms — strip after LLM processes (#648)
- Fix phantom action hallucination — detect when LLM claims completion without calling tools (#688)
- Fix OpenRouter model prefix not stripped on fallback chains (#630)
- Fix streaming path missing metering.record() — token usage now persisted (#627)
- Fix MCP server names with dashes not resolving (#616)
- Fix deepseek-reasoner multi-turn conversations (#618)
- Fix silent/NO_REPLY responses leaking to channels (#614)
- Fix installed skills still showing Install button (#625)
- Fix cron channel delivery was a no-op — all 3 delivery variants work (#601)

Features:
- Azure OpenAI provider with deployment-based URLs and api-key header (#631)
- LaTeX rendering in chat via KaTeX (#622)
- Progressive Web App support — manifest.json + service worker (#621)
- WeCom (WeChat Work) channel adapter with AES-256-CBC encryption (#629)
- Shell/Bash skill runtime (#624)
- DingTalk Stream mode adapter (#353)
- Feishu/Lark unified adapter with region toggle (#329)
- Stable hand agent IDs via UUID v5 (#520)
- Telegram reply-to-message context (#567)
- Slack app_mention event support (#540)
- Suppress error responses on broadcast channels (#536)
- Dashboard provider editing (#600)
- Codex GPT-5.4 model catalog (#608)
- Chromium --no-sandbox for root (#394)
- Tool error guidance to prevent fabricated results (#424)
- Channel agent re-resolution by name (#626)
- Slack unfurl_links config (#623)
- Docker build args for faster dev builds (#541)
- Parakeet MLX local speech-to-text (#607)
- OnceLock for peer registry (removes unsafe) (#526 partial)

Community:
- 100+ PRs reviewed and processed
- Telegram markdown block-level parser (#595)
- WhatsApp group reply routing (#604)
- Mastodon notification polling fix (#538)
- Async session save for 1-CPU deployments (#534)