Feature: Persistent cross-project memory via Dakera — agents recall prior related work

ChatDev 2.0 is an excellent multi-agent platform, but agents re-derive the same architectural decisions from scratch on every new task. This proposes Dakera (https://dakera.ai) as a persistent memory layer so agents can recall patterns from prior related projects.

Problem: When users give ChatDev similar tasks repeatedly, agents re-derive the same architectural decisions every time. A PM agent doesn't remember that last month's REST API project used FastAPI; an engineer doesn't recall the testing patterns used in a similar project.

Proposed Integration:
At ChatChain start, recall(task_description) retrieves memories from similar prior projects. At ChatChain end, store_memory(project_summary) persists key decisions.

In chat_chain.py:

• Before chain execution: inject recalled memories into initial system context
• After chain execution: store a summary of decisions made

Setup: docker run -d -p 3300:3300 -e DAKERA_API_KEY=demo ghcr.io/dakera-ai/dakera:latest / pip install dakera

Why Dakera fits ChatDev's philosophy: Both are fully self-hosted. Dakera's decay weighting means more recent successful project patterns surface preferentially — exactly what a simulated software company would want.

Happy to open a PR.

Start with chat_chain.py and review where chain execution begins and ends. Read Dakera's integration and setup details, including the docker command and pip install instructions, then define how recalled memories enter the initial system context and how a project summary is stored afterward. Done means the integration supports both operations without disrupting the existing chain flow.