Proper configuration for memorySearch using OpenRouter Embeddings
Hi everyone! I'm trying to optimize my OpenClaw setup to work efficiently with free-tier models (like Qwen 3 and GLM) while keeping long-term memory via RAG.
My goal is to keep my active context small (around 10k tokens) to avoid Rate Limits and high input costs, but still allow the agent to retrieve past information using Vector Memory Search.
I have $10 balance on OpenRouter and I want to use openai/text-embedding-3-small for the embeddings instead of a local model.
Could you verify if this agents.defaults configuration is correct for this use case?
"agents": {
"defaults": {
"memorySearch": {
"enabled": true,
"provider": "lancedb",
"embeddingModel": "openrouter/openai/text-embedding-3-small"
},
"compaction": {
"mode": "default",
"memoryFlush": {
"enabled": true,
"softThresholdTokens": 10000
}
},
"bootstrapMaxChars": 4000
}
}
Does OpenClaw support calling embeddingModel via the OpenRouter provider out of the box? When memoryFlush triggers, is the flushed content automatically indexed into lancedb for later memorySearch retrieval? Are there any other 'hidden' settings to ensure the primary agent prioritizes searching memory before failing due to truncated context?
Thanks in advance for the help! Any other tips to lower token usage would help!
My goal is to keep my active context small (around 10k tokens) to avoid Rate Limits and high input costs, but still allow the agent to retrieve past information using Vector Memory Search.
I have $10 balance on OpenRouter and I want to use openai/text-embedding-3-small for the embeddings instead of a local model.
Could you verify if this agents.defaults configuration is correct for this use case?
"agents": {
"defaults": {
"memorySearch": {
"enabled": true,
"provider": "lancedb",
"embeddingModel": "openrouter/openai/text-embedding-3-small"
},
"compaction": {
"mode": "default",
"memoryFlush": {
"enabled": true,
"softThresholdTokens": 10000
}
},
"bootstrapMaxChars": 4000
}
}
Does OpenClaw support calling embeddingModel via the OpenRouter provider out of the box? When memoryFlush triggers, is the flushed content automatically indexed into lancedb for later memorySearch retrieval? Are there any other 'hidden' settings to ensure the primary agent prioritizes searching memory before failing due to truncated context?
Thanks in advance for the help! Any other tips to lower token usage would help!
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
131,281Members
Resources
Similar Threads
Was this page helpful?
