Building a Local Multi-Agent Systems Lab
Living technical documentation for a local multi-agent environment — orchestration, code execution, research workflows, media automation, and system architecture.
What the lab is
The lab is a local-first personal environment for experimenting with agentic AI workflows. It supports software development, research automation, media management, and personal knowledge organization through a set of specialized agents that operate under defined boundaries.
This documentation records the architecture, implementation decisions, experiments, and design evolution of these systems. Pages are updated as the systems change.
Status: In progress. Core architecture is defined and partially implemented. Documentation is being written.
How this section is organized
Lab Notes is one of two writing formats on this site:
- Research Notes — editorial analysis of external systems, published architectures, companies, and AI governance topics.
- Lab Notes — living technical documentation for systems under active development and operation.
Lab Notes follows technical documentation conventions, not essay format.
System topology
Primary agents
Orchestration Agent
Status: Implemented (concept and primary development workflow)
The Orchestration Agent is the main planning and coordination component. It designs solutions, delegates implementation to the Code Execution Agent, delegates research to the Research Agent, and delegates media requests to the Media Agent. It does not execute code directly.
→ Orchestration Agent · Code Execution Workflow
Code Execution Agent
Status: In use as a controlled implementation layer
The Code Execution Agent handles code generation and modification under explicit workflow constraints. It does not interact with git or production systems directly.
Research Agent
Status: Implemented (task queue, mission runner, phase DAG, and artifact pipeline)
The Research Agent executes structured research missions in the background through a file-based task queue, multi-phase pipeline, and typed artifact store.
→ Research Agent · Research Mission DAG · Research Artifacts
Media Agent
Status: Implemented (media search, streaming integration, and playback control)
The Media Agent handles local entertainment automation — media search, platform availability evaluation, and playback initiation to connected devices.
Navigation
| Section | Purpose |
|---|---|
| System Overview | High-level topology and component roles |
| Data Flow | How requests, tasks, and artifacts move through the system |
| Orchestration Agent | Planning, coordination, and memory |
| Code Execution Agent | Implementation workflow and constraints |
| Research Agent | Structured background research |
| Research Framework | Multi-phase research pipeline |
| Research Mission DAG | Phase dependency graph |
| Research Artifacts | Mission output files and lifecycle |
| Media Agent | Entertainment automation |
| Roadmap | Implemented, planned, and future work |