Memory track
This track teaches how ArcFlow agents attach memory backends through MemoryConfig. You start with the four memory types and scope rules, then practice session and shared memory without external services, and finish with vector memory setup for Qdrant-backed RAG.
What you will learn
| Lesson | Topic |
|---|---|
| 01 Memory types overview | MemoryType, MemoryScope, and when to pick each combination |
| 02 Session and shared memory | In-run scratch state and multi-agent handoff |
| 03 Vector memory setup | Qdrant via ARCFLOW_QDRANT_URL, namespace, stub embeddings |
Prerequisites
Complete Install and build so from arcflow import Agent, MemoryConfig, MemoryScope, MemoryType, Workflow works in your virtual environment. Read 03 Anatomy of a workflow and Writing agents so agent construction and step wiring are already familiar.
Lessons 01 and 02 run with the stub provider and no Docker. Lesson 03 adds Qdrant for vector storage; stub embedding (stub/8 or stub/384) keeps local runs working without a real embedding API key.
How these lessons are structured
Every page follows the same sections: Before you start, Concept, Example, Verify, and Next. Run each example as a standalone script unless the page says otherwise.
After this track
| Goal | Next document |
|---|---|
| Ingest documents and wire RAG agents | RAG track |
| Guided verification with trace events | Track C: RAG and vector memory |
| Full memory reference | Memory types |