没有openclaw embedding 配置功能

[csuwl]

Pre-checks

• I have searched existing issues and this is not a duplicate.

What problem does this solve?

可以新增一个embedding配置功能，开启embedding记忆搜索功能

Proposed solution

No response

Willing to contribute?

• I'd be interested in working on this.

[Clawiee]

Hi @csuwl 👋

Thanks for submitting this feature request!

Analysis

I reviewed the Clawith codebase and found that the current memory system works as follows:

Current Memory Architecture:

• Agents store long-term memory in memory/memory.md files (agent_context.py)
• Memory is loaded as plain text and injected into the agent's system prompt (up to 2000 chars)
• This is a token-efficient approach but doesn't support semantic search

What's Missing for Embedding-Based Memory Search:

To implement embedding-based memory retrieval, the system would need:

1. Vector Database Integration — Store memory embeddings (e.g., via Pinecone, Weaviate, or local FAISS)
2. Embedding Model Configuration — Allow users to select/configure embedding models (OpenAI, Ollama, etc.)
3. Memory Indexing Pipeline — Automatically embed memory chunks when they're created/updated
4. Semantic Search at Runtime — Retrieve relevant memory chunks based on query embeddings instead of loading all memory
5. UI Configuration Panel — Let users enable embedding search and configure:
   • Embedding model selection
   • Vector DB connection details
   • Chunk size and overlap parameters
   • Similarity threshold for retrieval

Proposed Implementation Path:

1. Phase 1 (Backend):

   • Add embedding_config field to the Agent model (model name, vector DB type, enabled flag)
   • Create an EmbeddingService to handle embedding generation and storage
   • Modify memory loading to optionally retrieve chunks via semantic search
   • Add embedding indexing to the memory update workflow

2. Phase 2 (Frontend):

   • Add "Memory Settings" tab in Agent configuration
   • Embedding toggle + model selection dropdown
   • Vector DB connection form (if self-hosted)

3. Phase 3 (UX):

   • Display memory retrieval confidence scores in agent chat
   • Allow users to refresh/rebuild embedding index

Potential Challenges:

• Cost implications (embedding API calls)
• Latency impact on agent responses
• Storage overhead for vector embeddings
• Need for graceful fallback if embedding service is unavailable

---

This analysis is based on reviewing the Clawith source code. I recommend discussing this with the maintainers to confirm the implementation approach and prioritization. 🙏

[wisdomqin]

Adding vector search would increase the project's complexity and raise deployment requirements due to the need for a vector database. However, the benefits don't seem particularly significant, so the ROI feels low. We won't be supporting this in the short term for now. We'll continue to monitor the situation going forward, and you're welcome to keep sharing any issues you encounter due to the lack of vector search.