Skip to content
Draft
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 29 additions & 0 deletions automations/livekit-agent/2026-08-08.markdown
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
# LiveKit Agents 工程日报

**Date range covered:** 2026-08-07 00:01 UTC → 2026-08-08 00:01 UTC

**仓库:** [livekit/agents](https://github.com/livekit/agents)(upstream/main)

---

## 关键变更

- **PyPI 发布 livekit-agents@1.6.9** — [#6753](https://github.com/livekit/agents/pull/6753) 将全量包版本升至 **1.6.9**(自 1.6.8 起累计 **36** 个 commit)。发布窗口内合入的修复/特性包括:TTS fallback 流式恢复 ([#6684](https://github.com/livekit/agents/pull/6684))、Deepgram `end_time` 末词修正 ([#6712](https://github.com/livekit/agents/pull/6712))、`session_host` ([#6725](https://github.com/livekit/agents/pull/6725))、Baseten Qwen3 STT/TTS ([#6700](https://github.com/livekit/agents/pull/6700))、OpenAI STT context hints + `openai>=2.50` ([#6705](https://github.com/livekit/agents/pull/6705))、Fish quality-guard ([#6734](https://github.com/livekit/agents/pull/6734))、Nova Sonic 音频门控 ([#6720](https://github.com/livekit/agents/pull/6720))、IPC PSS 内存报告 ([#6082](https://github.com/livekit/agents/pull/6082)) 与 `is_alive` 修复 ([#6732](https://github.com/livekit/agents/pull/6732)) 等。影响:生产升级需关注 voice/TTS/realtime 行为变化。

- **Expressive 模式正式对外 + 示例** — [#6698](https://github.com/livekit/agents/pull/6698) 在公共 API 暴露 expressive 模式,新增 `_mood.py` / `_mood_data.py`(+445 行)与 `examples/expressive_agent/` 完整示例;`drive-thru` 目录重命名为 `drive_thru` 并更新 CI/deploy 配置。影响:开发者可直接启用情绪/表现力 TTS 控制;示例路径变更需同步部署脚本。

- **xAI Realtime 每轮仅一条最终转写** — [#6729](https://github.com/livekit/agents/pull/6729) 修复 xAI 在每次 audio commit 重复发送 `input_audio_transcription.completed` 导致同一轮出现多条 `user_input_transcribed`、历史重复与客户端分段的问题;同时修复 OpenAI realtime 插件中 dropped response 流未关闭、rejected item 阻塞 `update_chat_ctx` 等路径(+336 行 xAI 测试、+125 行 OpenAI 测试),并新增 `grok-voice-think-fast-2.0` 模型。影响:xAI/OpenAI realtime 会话转写与打断可靠性显著提升。

- **工具重复调用策略细化** — [#6610](https://github.com/livekit/agents/pull/6610) 为 `@function_tool` 新增 `duplicate_scope`(`"name"` / `"name_and_args"`),允许按工具名或「工具名+参数」判定重复;`tool_executor.py` 实现规范化参数比较,`tests/test_tools.py` 新增 **+262** 行覆盖。影响:并发/重复工具调用场景可更精细地 reject/replace/confirm,但文档明确「fail open」——不可作为 exactly-once 保证。

- **Inference 429 配额可观测性** — [#6736](https://github.com/livekit/agents/pull/6736) 在 `inference.LLM` 收到 429 时解析 gateway 的 `X-LiveKit-Inference-*` 头(RPM/TPM/Credits limit & used),输出结构化 warning 后再抛出 `APIStatusError`;`tests/test_inference_utils.py` 新增 11 项单测。影响:限流排障可直接看到触发的配额维度,旧 gateway 仅记录 model + request_id。

---

## Watchlist

1. **Voice 生命周期修复栈仍未合入** — [#6733](https://github.com/livekit/agents/pull/6733)(用户说话时禁止启动回复 playout)、[#6714](https://github.com/livekit/agents/pull/6714)(待转写时延迟 false-interruption resume)、[#6745](https://github.com/livekit/agents/pull/6745)(playout 启动时挂起 pending replies)均 OPEN;与已合入的 [#6729](https://github.com/livekit/agents/pull/6729) xAI 修复存在重叠关注面,realtime 打断/转写回归风险仍在。

2. **Expressive 后续与 xAI 默认值对齐** — [#6747](https://github.com/livekit/agents/pull/6747)(自定义 expressiveness steering)在 [#6698](https://github.com/livekit/agents/pull/6698) 合入后同日更新;[#6755](https://github.com/livekit/agents/pull/6755)(xAI realtime Voice Agent 默认值与 API 覆盖对齐)为 [#6729](https://github.com/livekit/agents/pull/6729) 的跟进,升级 1.6.9 后建议关注 xAI 集成是否需同步配置。

3. **Fork 落后 upstream 408 commits** — `origin/main` 未同步 1.6.9 及本周 voice/inference/expressive 变更;另 [#6739](https://github.com/livekit/agents/issues/6739) 报告 Cartesia WebSocket 握手失败时 API Key 泄漏至日志,对应修复 [#6740](https://github.com/livekit/agents/pull/6740) 仍 OPEN,生产日志需留意敏感信息。