weave-search

name	weave-search
description	Query vector databases and manage AI agents using weave-cli. Use when: (1) running semantic, BM25, or hybrid searches, (2) creating and configuring RAG/QA/summarize agents, (3) comparing embedding models, (4) re-embedding collections, (5) tuning search parameters. NOT for: initial setup (use weave-setup), data ingestion (use weave-ingest), evaluations (use weave-eval), or deployment (use weave-stack).
homepage	https://github.com/Maximilien-ai/weave-cli
tags	["rag","data","weave","vector-database"]
metadata	{"openclaw":{"emoji":"🧠","requires":{"bins":["weave"]},"install":[{"id":"brew","kind":"brew","formula":"Maximilien-ai/tap/weave-cli","bins":["weave"],"label":"Install weave-cli (brew)"},{"id":"go","kind":"go","package":"github.com/Maximilien-ai/weave-cli","bins":["weave"],"label":"Install weave-cli (go install)"}]}}

weave-search

Query vector databases and manage AI agents for RAG, QA, and summarization.

When to Use

• Running semantic, BM25, or hybrid searches against collections
• Creating and managing weave AI agents (RAG, QA, summarize, custom)
• Comparing embedding models across collections
• Re-embedding collections with different models
• Tuning search relevance and parameters

When NOT to Use

• Initial setup and config -> weave-setup skill
• Data ingestion -> weave-ingest skill
• Evaluations and benchmarks -> weave-eval skill
• Stack deployment -> weave-stack skill

Querying

Direct Collection Query

# Semantic search (default)
weave collection query MyDocs "what is the return policy?"
weave collection query MyDocs "find documents about pricing" --top-k 10

# Shorthand
weave query MyDocs "search terms"

Search Modes

# Semantic (vector similarity)
weave collection query MyDocs "query" --mode semantic

# BM25 (keyword matching)
weave collection query MyDocs "query" --mode bm25

# Hybrid (semantic + BM25 combined)
weave collection query MyDocs "query" --mode hybrid

Metadata Filtering

weave collection query MyDocs "query" \
  --filter '{"category": "pricing", "year": 2026}'

AI Agents

Built-in Agent Types

Type	Temperature	Best For
rag	0.7	General document Q&A, research
qa	0.3	Precise answers, customer support, compliance
summarize	0.5	Document summaries, bullet-point overviews
custom	User-defined	Specialized use cases

Agent Management

# Create agent
weave agents create my-rag --type rag
weave agents create my-qa --type qa
weave agents create my-summarizer --type summarize
weave agents create custom-agent --type custom

# List agents
weave agents list

# Show agent config
weave agents show my-rag

# Edit agent config
weave agents edit my-rag

# Copy agent
weave agents copy my-rag my-rag-v2

# Delete agent
weave agents delete my-rag

Agent Configuration (weave-agents.yaml)

agents:
  - name: auction-rag
    type: rag
    model: gpt-4o
    temperature: 0.1
    max_tokens: 2000
    collections:
      - AuctionListings
    system_prompt: |
      You are an auction catalog expert.
      Always cite sources with [1], [2] notation.

Re-embedding & Comparison

Re-embed Collection (20x faster than re-ingestion)

# Re-embed with OSS model
weave collection re-embed MyDocs \
  --new-embedding sentence-transformers/all-mpnet-base-v2 \
  --output MyDocs_OSS

Compare Embedding Models

# Side-by-side comparison
weave collection compare MyDocs MyDocs_OSS \
  --query "search terms" \
  --report comparison.md

Search Tuning

Top-K Results

weave query MyDocs "query" --top-k 5    # Fewer, more relevant
weave query MyDocs "query" --top-k 20   # More results, broader

JSON Output

weave query MyDocs "query" --json

Parallel Search Pattern (for ClawMax multiagent)

Multiple search agents can test different query strategies concurrently:

# Agent 1: Semantic search tuning
weave query MyDocs "query" --mode semantic --top-k 5

# Agent 2: Hybrid search tuning
weave query MyDocs "query" --mode hybrid --top-k 10

# Agent 3: Different agent type
weave agents create qa-strict --type qa

Decision Workflow

1. Start with semantic search: weave query MyDocs "test query"
2. Try hybrid if keyword precision needed
3. Create appropriate agent type (RAG for general, QA for precise)
4. Compare embedding models if quality isn't sufficient
5. Re-embed with better model if needed
6. Hand off to weave-eval for systematic quality measurement