OG T
|
797c7c749e
|
fix(nemotron): deepseek-r1 num_predict 400→1200,避免 <think> block 截斷後空回覆
CD Pipeline / build-and-deploy (push) Failing after 28s
deepseek-r1:14b 思考 token 超過 400 會在 </think> 前截斷,導致
清理後 body 為空,Telegram 顯示空訊息。
- chat_manager: num_predict 400 → 1200
- telegram_gateway: _clean_ai_reply 空值加 fallback 錯誤提示
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-10 22:35:37 +08:00 |
|
OG T
|
100e4d9b89
|
fix(chat): AI 回覆截斷問題 — 強制 persona + Markdown 清理 + 600字上限
CD Pipeline / build-and-deploy (push) Successful in 14m39s
問題: OpenClaw/NemoClaw 回覆 Markdown 語法 + 超長,Telegram 顯示截斷
修正:
1. chat_manager: _call_openclaw/_call_nemotron 強制前置 persona (含不超過300字規範)
2. telegram_gateway: _clean_ai_reply() 移除 **bold** *italic* # header 語法
移除 deepseek-r1 <think> 標籤,截斷 > 600 字並在段落邊界截
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-10 21:26:15 +08:00 |
|
OG T
|
a30713b292
|
fix(chat): NemoClaw 禁止自稱 DeepSeek + 強制繁體中文
CD Pipeline / build-and-deploy (push) Successful in 13m36s
- 明確禁止透露底層模型身分
- 強制繁體中文(禁簡體)
- 補充 SRE 專長範圍定義
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-10 01:18:18 +08:00 |
|
OG T
|
a303b5ef91
|
feat(chat): NemoClaw 改接 Ollama 111 deepseek-r1:14b
CD Pipeline / build-and-deploy (push) Failing after 4m6s
2026-04-09 ogt: 棄用 Claude Haiku,改用本地 deepseek-r1:14b
- 端點: http://192.168.0.111:11434
- 過濾 <think>...</think> 推理區塊,只回傳結論
- timeout 120s(14b 推理較慢)
- 完全免費,不計入 Claude API 費用
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-09 23:38:57 +08:00 |
|
OG T
|
b62d7d3eb0
|
feat(chat): OpenClaw 改用 Gemini 2.0 Flash-Lite (最便宜)
CD Pipeline / build-and-deploy (push) Has been cancelled
Input $0.075/1M, Output $0.30/1M (比 Flash 便宜 25%)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 19:35:13 +08:00 |
|
OG T
|
6cd4280168
|
feat(chat): NemoClaw Claude API 加 token+費用統計
CD Pipeline / build-and-deploy (push) Has been cancelled
Claude Haiku 4.5: Input $0.80/1M, Output $4.00/1M
每次回覆顯示: token 數 | 本次費用 | 本月累計
Redis key: claude_cost:YYYY-MM,TTL 40 天
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 19:29:22 +08:00 |
|
OG T
|
781a6dac3e
|
feat(chat): NemoClaw→Claude Haiku API + 告警只由 OpenClaw 分析
CD Pipeline / build-and-deploy (push) Successful in 7m20s
老闆指示 (2026-04-03):
1. NemoClaw 改接 Claude API (claude-haiku-4-5),快速中文對話
2. 群組告警分析只觸發 OpenClaw,NemoClaw 不分析告警
3. OpenClaw/NemoClaw 雙向自然語言對話維持
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 19:19:56 +08:00 |
|
OG T
|
10ad2a67c7
|
fix(chat): gemini-2.0-flash 修正 + 全形小O支援 + NemoClaw 回 NIM
CD Pipeline / build-and-deploy (push) Has been cancelled
1. Gemini 模型名稱: gemini-1.5-flash → gemini-2.0-flash (404修復)
2. 費用計算: 2.0 Flash 定價 Input $0.10/1M, Output $0.40/1M
3. 全形/半形統一: unicodedata.normalize NFKC,支援「小O」全形輸入
4. NemoClaw: Ollama 188 負載高超時,暫回 NIM nemotron-mini-4b
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 19:17:08 +08:00 |
|
OG T
|
08b02280f8
|
feat(chat): Gemini 月費用上限 $10 USD + Redis 累計追蹤
CD Pipeline / build-and-deploy (push) Successful in 6m55s
- 每次呼叫前檢查當月累計費用,超過 $10 USD 拒絕呼叫
- Redis key: gemini_cost:YYYY-MM,TTL 40 天
- 每次回覆顯示: token 數 | 本次費用 | 本月累計
- 超限時回傳警告訊息告知老闆
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 19:01:21 +08:00 |
|
OG T
|
2828cd897a
|
feat(chat): OpenClaw→Gemini Flash + NemoClaw→Ollama llama3.2:3b
CD Pipeline / build-and-deploy (push) Has been cancelled
老闆指示 (2026-04-03):
- OpenClaw: Gemini 1.5 Flash API,每次回覆附 token+費用統計
- NemoClaw: Ollama llama3.2:3b,本地快速回應 (3-8s)
- 費用控管: Gemini 月上限 $10 USD
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 18:59:28 +08:00 |
|
OG T
|
fbf122fa1f
|
fix(chat): OpenClaw 改用 NIM llama-3.1-8b 對話 + NemoClaw timeout 120s + 老闆稱謂
CD Pipeline / build-and-deploy (push) Successful in 7m9s
1. _call_openclaw: 改用 NIM meta/llama-3.1-8b-instruct
舊的 analyze/incident 是告警 API,回覆是告警格式,不適合對話
2. _call_nemotron: 移除 Ollama fallback,回到純 NIM
3. NEMOTRON_TIMEOUT_SECONDS: 55 → 120 (ConfigMap 已更新)
4. 修正「統帥」→「老闆」
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 18:41:15 +08:00 |
|
OG T
|
2da8da5a25
|
fix(chat): OpenClaw 改用 Ollama qwen2.5 做對話 + NemoClaw 加 Ollama fallback
CD Pipeline / build-and-deploy (push) Successful in 6m51s
問題: _call_openclaw 用 analyze/incident API → 回覆是告警格式,不是自然語言
修法:
1. OpenClaw chat → Ollama qwen2.5:7b-instruct (本地,快速,無格式污染)
2. NemoClaw → NIM 優先,超時 fallback 到 Ollama llama3.2:3b
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 18:30:31 +08:00 |
|
OG T
|
63929a5e87
|
feat(group): 別名 小O→OpenClaw 小賀→NemoClaw + NemoClaw 強制繁中
CD Pipeline / build-and-deploy (push) Successful in 7m6s
1. telegram_gateway.py: _handle_group_message 加入別名路由
- 小O / 小o → 只有 OpenClaw 回應
- 小賀 / 小贺 → 只有 NemoClaw 回應
- clean_text 同步移除別名 token
2. chat_manager.py: NEMOCLAW_PERSONA 加強繁體中文強制指令
- 明確「禁止使用英文或其他語言」防止 Nemotron 自動英文回應
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 18:00:51 +08:00 |
|
OG T
|
699e61ac87
|
feat(group): 群組雙向對話 + 格式選項C + 老闆稱謂
CD Pipeline / build-and-deploy (push) Successful in 7m11s
1. _handle_group_message: SRE 群組訊息路由
- @OpenClawAwoooI_Bot → 只有 OpenClaw 回應
- @NemoTronAwoooI_Bot → 只有 NemoClaw 回應
- 一般訊息 → 並行回應 + 互相評論第二輪
- Bot 訊息自動忽略(防無限循環)
2. 告警格式改選項 C (老闆指示)
- 【🔴 HIGH】resource_name
- 區塊式,去掉 ═══ 長分隔線
3. AI persona 改稱呼用戶為「老闆」
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 17:51:48 +08:00 |
|
OG T
|
15aabd6ac5
|
fix(chat+nim): 修復首席架構師 Review I1-I4 + S3 四項重要問題
CD Pipeline / build-and-deploy (push) Successful in 7m9s
I1: chat_manager._call_openclaw timeout=30.0 → 讀 settings.OPENCLAW_TIMEOUT
I2: nvidia_provider.py stale comment "45" → "55" 對齊 ConfigMap
I3: asyncio.shield 移除 — shield 超時後 task 繼續跑但無人等待 (silent leak)
I4: ChatManager.__init__ 移除 repo 實例 (leWOOOgo 禁 Service 持有 repository)
S3: _check_nemotron_health probe 10s → 25s + /v1/models 輕量端點
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 16:36:16 +08:00 |
|
OG T
|
be247d6c5c
|
fix(chat): OpenClaw timeout 30→40s,NemoClaw 50→60s
CD Pipeline / build-and-deploy (push) Successful in 6m51s
get_system_context() k8s/DB 查詢加上 _call_openclaw 30s,
總計超過外層 shield 30s 導致 OpenClaw 全部超時。
放寬 timeout 讓兩個 AI 有足夠時間回應。
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 16:27:08 +08:00 |
|
OG T
|
1430b1283d
|
fix(chat+nvidia): 還原 OpenClaw+Nemotron 架構 + 修 30s timeout 根因
CD Pipeline / build-and-deploy (push) Has been cancelled
ChatManager 還原:
- OpenClaw (188:8088) 負責 RCA 仲裁,不改用 Gemini (未經批准)
- NemoClaw (NVIDIA NIM nemotron-mini-4b) 負責補充/評論
- 雙 AI 並行執行,OpenClaw 30s / NemoClaw 50s timeout
- 支援 @openclaw / @nemo 指定對象
nvidia_provider.py 修 timeout 根因:
- NVIDIA_TIMEOUT 從硬編碼 30.0 改為讀 NEMOTRON_TIMEOUT_SECONDS (45s)
- Memory 記載 NIM 免費 tier 延遲 11-45s,30s 硬編碼導致慢請求全超時
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 15:34:02 +08:00 |
|
OG T
|
e93ada0452
|
fix(chat): OpenClaw 改走 Gemini Flash,移除 Ollama 依賴
CD Pipeline / build-and-deploy (push) Successful in 7m18s
Ollama 188 完全卡死 (0 bytes/30s timeout),無法作為對話後端。
雙 AI 皆使用 Gemini Flash,靠不同 persona 和 temperature 區分:
- OpenClaw: temperature=0.5 (精準果斷)
- NemoClaw: temperature=0.9 (分析發散)
同時 kubectl set env ENABLE_NEMOTRON_COLLABORATION=false
停止每個 incident 白白等待 30s Nemotron timeout
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 15:20:23 +08:00 |
|
OG T
|
d9007e6855
|
feat(chat+monitor): 雙 AI 對話重寫 + Nemotron 健康監控告警
CD Pipeline / build-and-deploy (push) Successful in 6m56s
ChatManager 重寫 (Phase 22.6):
- @openclaw <msg> → 只有 OpenClaw 回應 (Ollama qwen2.5:7b)
- @nemo <msg> → 只有 NemoClaw 回應 (Gemini Flash)
- 無前綴 → OpenClaw 先答,NemoClaw 評論/反駁
NemoClaw 改用 Gemini Flash (棄 NIM nemotron-mini-4b 因為 15s+ 回應時間)
TelegramGateway 心跳新增 Nemotron 健康探測:
- 每次心跳探測 NVIDIA NIM API (10s timeout)
- 異常時立刻發 Telegram 告警 + 緩解指令
- 補足 Nemotron 100% 超時卻無告警的監控盲區
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
2026-04-03 14:59:06 +08:00 |
|
OG T
|
8b7f99b5fa
|
fix(telegram): fix chat_id routing and llm result unpacking
E2E Health Check / e2e-health (push) Successful in 18s
|
2026-03-31 15:56:58 +08:00 |
|
OG T
|
3260c565ef
|
feat(telegram): enable interactive chat with Nemo-4B context
E2E Health Check / e2e-health (push) Successful in 16s
|
2026-03-31 15:44:49 +08:00 |
|