Documentation
Get Started

Vector stores

Use Pinecone, Qdrant, Weaviate, Chroma, Upstash Vector, and Zilliz in Action nodes and AI agent memory.

TogoFlow connects to popular vector databases for RAG-style retrieval and AI agent memory. Load and manage documents in your vector provider (or with your own pipeline); TogoFlow searches and upserts vectors in workflows.

Supported stores

AppBest for
PineconeManaged serverless / pod indexes
QdrantCloud or self-hosted collections
WeaviateCloud or self-hosted class-based search
ChromaCloud or self-hosted collections
Upstash VectorServerless REST indexes
Zilliz / MilvusZilliz Cloud (Milvus) clusters

Actions

ActionUse
Similarity SearchNearest-neighbor query — link as AI agent memory read action
UpsertWrite vectors you already prepared (ids + embeddings + metadata)
DeleteRemove by ids or filter

TogoFlow does not run a full document ingest pipeline (no automatic file chunking / knowledge-base upload UI). Put knowledge into the index with your provider’s tools, SDK, or the Upsert action when you already have vectors.

For search/upsert you can pass a precomputed vector, or optionally text to embed via OpenAI when you want a one-off query embedding.

Quick path (copy-paste)

  1. Load docs into your provider (example with OpenAI embeddings + Pinecone SDK, or the provider console).
  2. In TogoFlow: Connections → your vector app → paste credentials → Test connection.
  3. Add Similarity Search as an Action, or AI agent → Add memory → pick the store (configures search on the memory chip — no extra canvas node).
  4. Set search text to:
{{inputs.message}}
  1. Pick namespace/collection from the dropdown (or type it). Optionally set Expected dimensions (e.g. 1536).
  2. Run — ranked search hits hydrate agent memory (and show in the node data viewer when using a canvas Action).

Templates

TemplateWhat it does
Pinecone RAG ChatChat → AI with Pinecone memory chip (+ setup checklist)
Qdrant RAG ChatSame pattern with Qdrant

Load docs into the index first, then use the template. Follow the amber sticky notes on the canvas.

Action node

  1. Add an Action → vector store → connect credentials (test connection).
  2. Use Similarity Search with text or vector (default {{inputs.message}}).
  3. Open Advanced for raw vector, embedding model/key, and metadata filters.
  4. Downstream nodes read {{lastOutput}} / {{lastOutput.rows}}.

AI agent memory

  1. AI node (agent mode) → Add memory → pick Pinecone / Qdrant / etc.
  2. Configure credentials and search fields on the memory chip (test connection).
  3. Publish — the agent runs similarity search at the start of the run and injects hits into context.
  4. Optionally attach Upsert as a tool if the agent should write vectors you supply in the tool call.

Related