Search

Search your stored conversations using semantic similarity. Find relevant sessions by meaning, not just exact keywords.

Prerequisites

SQLite Database For storing conversation sessions

Vector Database SQLite, Chroma, Qdrant, or pgvector for storing embeddings

Ollama with Embedding Model For generating embeddings from conversations

Search Guide

Step 1: Choose your vector store

tapes supports 4 vector backends. Pick one:

SQLite VectorChromaQdrantpgvector

Built-in, no extra setup. Uses the same SQLite database as your conversation storage.

External vector store. Start it with Docker:

docker run -p 8000:8000 chromadb/chroma

High-performance vector store with gRPC API. Start it with Docker:

docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

Dashboard available at http://localhost:6333/dashboard

PostgreSQL extension for vector similarity search. Requires PostgreSQL with the pgvector extension installed:

docker run -p 5432:5432 -e POSTGRES_PASSWORD=postgres pgvector/pgvector:pg17

Uses HNSW indexing for efficient cosine similarity queries.

Step 2: Install Ollama embedding model

Download the embedding model that converts text into searchable vectors:

ollama pull nomic-embed-text

Step 3: Run proxy with vector storage enabled

Start the tapes proxy with your chosen vector backend:

SQLite VectorChromaQdrantpgvector

tapes serve \
  --sqlite "./tapes.sqlite" \
  --vector-store-provider sqlite \
  --embedding-provider ollama \
  --embedding-target "http://localhost:11434" \
  --embedding-model nomic-embed-text

tapes serve \
  --sqlite "./tapes.sqlite" \
  --vector-store-provider chroma \
  --vector-store-target "http://localhost:8000" \
  --embedding-provider ollama \
  --embedding-target "http://localhost:11434" \
  --embedding-model nomic-embed-text

tapes serve \
  --sqlite "./tapes.sqlite" \
  --vector-store-provider qdrant \
  --vector-store-target "http://localhost:6334" \
  --embedding-provider ollama \
  --embedding-target "http://localhost:11434" \
  --embedding-model nomic-embed-text

tapes serve \
  --sqlite "./tapes.sqlite" \
  --vector-store-provider pgvector \
  --vector-store-target "postgres://postgres:postgres@localhost:5432/postgres" \
  --embedding-provider ollama \
  --embedding-target "http://localhost:11434" \
  --embedding-model nomic-embed-text

The tapes serve command automatically targets Ollama on localhost. The flags above enable vector storage for semantic search.

Step 4: Use your LLM through the proxy

Embeddings are generated automatically when vector storage is enabled:

curl -X POST http://localhost:8080/api/chat \
  -H "Content-Type: application/json" \
  -d '{"model": "qwen3", "messages": [{"role": "user", "content": "How do I configure logging?"}]}'

Step 5: Search your sessions

Query your conversation history using natural language. Use the same vector store provider you chose above:

SQLite VectorChromaQdrantpgvector

tapes search "logging configuration" \
  --vector-store-provider sqlite \
  --embedding-provider ollama \
  --embedding-target http://localhost:11434 \
  --embedding-model nomic-embed-text \
  --sqlite ./tapes.sqlite

tapes search "logging configuration" \
  --vector-store-provider chroma \
  --vector-store-target http://localhost:8000 \
  --embedding-provider ollama \
  --embedding-target http://localhost:11434 \
  --embedding-model nomic-embed-text \
  --sqlite ./tapes.sqlite

tapes search "logging configuration" \
  --vector-store-provider qdrant \
  --vector-store-target http://localhost:6334 \
  --embedding-provider ollama \
  --embedding-target http://localhost:11434 \
  --embedding-model nomic-embed-text \
  --sqlite ./tapes.sqlite

tapes search "logging configuration" \
  --vector-store-provider pgvector \
  --vector-store-target "postgres://postgres:postgres@localhost:5432/postgres" \
  --embedding-provider ollama \
  --embedding-target http://localhost:11434 \
  --embedding-model nomic-embed-text \
  --sqlite ./tapes.sqlite

Understanding Results

Search results include:

Score Similarity score (higher = more relevant)

Hash Unique identifier for the matched conversation turn

Session ancestry Full conversation thread from root to matched message

Preview Snippet of the matched content

Next Steps

Once you've found a relevant conversation, use the checkout command to resume from that point. See the History guide for details on managing session history and resuming conversations.