新手不再“失忆”：把 memorySearch embedding 配置纳入首装必检
场景：完成 setup 但未配置 embedding，导致 memory_search 无结果、跨会话记忆失效。前置：有可用 embedding provider（Gemini/OpenAI 等）与 key。步骤：安装后立即补齐 memorySearch provider+model+密钥，再做测试查询。关键配置：memorySearch.provider/model。验证：memory_search 可返回历史片段。风险：把问题误判为模型质量。
GITHUBDiscovered 2026-02-15Author actwebmaster
Prerequisites
- You have an embedding-capable provider account and API key.
- Gateway config can be edited and service can be restarted safely.
Steps
- After initial setup, explicitly configure `memorySearch.provider` and embedding model.
- Inject provider API key through secure env/config path (avoid hardcoding in shared files).
- Restart gateway and run a small memory write + memory_search test query.
- If zero results, verify embedding endpoint, model name, and auth scope first.
Commands
openclaw gateway statusopenclaw gateway restartopenclaw helpVerify
memory_search returns semantically relevant snippets from newly written notes.
Caveats
- No embedding config means no durable recall, even if chat quality looks normal initially.
- Provider model naming and quota limits vary; validate against official docs before rollout（需验证）.
Source attribution
This tip is aggregated from community/public sources and preserved with attribution.
Open original source ↗