ChatDev + MemorySync
A memorysync store type for ChatDev’s zero-code platform: one pip install adds it to workflow YAML and the web console — semantic recall with budgets, honest timestamps, fact extraction from each node’s input.
Overview
Install chatdev-memorysync inside a ChatDev (DevAll) checkout and the memorysync store type joins the platform’s memory registry — usable from workflow YAML and the web console’s store-type dropdown, activated automatically by a .pth hook that no-ops outside ChatDev. The store is engineered against the in-tree Mem0 store’s failure modes: user_id is a REQUIRED config field (Mem0’s demo hardcodes a shared id so every operator shares one partition); recall runs under a hard 1.2s budget and fails open with status-bearing logs (Mem0’s SDK call has NO timeout — a slow backend stalls the whole agent turn — and its bare error logging makes quota, auth, and network failures indistinguishable); retrieved items carry the row’s real created_at (Mem0 stamps everything time.time(), corrupting ChatDev’s own time-decay scorer, which weights age at 0.7); after each agent node that writes to the store, the node’s input goes to fact extraction as the user’s turn and only the durable facts in it become memories (agent outputs are not sent, and store_outputs is accepted but has no effect); a deterministic seed lets the server recognise a re-sent input, so it is extracted once (Mem0 re-extracts duplicates). similarity_threshold is honored on real scores. Memories flow to and from every other MemorySync surface.
Setup status and requirements
- Supported version
- chatdev-memorysync 1.1.0 (PyPI)
- Last setup review
- 2026-09-01
- Permissions
- A MemorySync API key with read and write scopes.
- Limits
- Requires a ChatDev (DevAll) checkout (Python 3.12+).
Capabilities
- Config-driven: type: memorysync in YAML + the web console dropdown — zero code
- Automatic .pth activation; safe no-op outside ChatDev (subprocess-tested)
- user_id REQUIRED — no shared demo partitions
- Hard 1.2s recall budget, fail-open with status-bearing logs
- Honest created_at timestamps — plays fair with ChatDev’s decay scorer
- Each node’s input goes to fact extraction; agent outputs are not sent
- Deterministic idempotency seeds — a re-run input is extracted once
- similarity_threshold honored on real scores
Quick Start
graph:nodes:- id: writertype: agentconfig:memories:- name: user_memorytop_k: 5retrieve_stage: [gen]memory:- name: user_memorytype: memorysyncconfig:api_key: ${MEMORYSYNC_API_KEY}user_id: customer-42
Use Cases
ChatDev workflows that remember users across runs
Replacing the in-tree Mem0 store and its shared-partition demo
Cross-surface memory: facts learned in ChatDev recallable everywhere
Explore More
Built for production AI systems
Build AI systems that remember
MemorySync provides the infrastructure layer for persistent memory, adaptive retrieval, and enterprise AI intelligence.
Start free. No credit card required.