LIVE ARCHITECTURE · MULTI-AGENT SYSTEM

Architecture of my
multi-agent platform

41+ agents working in parallel on a shared code loop. Click any node to see details.

ALL SYSTEMS OPERATIONAL

// uptime

99.7%

// agents active

41+

// commits today

217

// throughput

150–300/d

ORCHESTRATOR

CODE-AGENT x12 writes · tests 12 / 12 active

REVIEW-AGENT x8 reviews · validates 6 / 8 active

TEST-AGENT x9 integration · e2e 9 / 9 active

MEMORY PostgreSQL + Redis + Qdrant

LLM (local) MCP servers Git / repos CI / Docker Observability

Orchestrator Code-agent Review-agent Test-agent Memory node Data streams

—

Role: —

Status: —

Total commits: —

Last activity: —

Which components participate in the work and what they're responsible for.

// orchOrchestrator

Central coordinator: distributes tasks across agents, gathers results, validates quality. Written in Python with async orchestration.

// codeCode-agent ×12

Write code, generate tests, refactor. Each one carries its own context and specialisation (frontend / backend / data).

// reviewReview-agent ×8

Review code from other agents, check style, hunt bugs, validate architectural decisions. 3-of-5 voting scheme.

// testTest-agent ×9

Run integration and e2e tests. Generate new test cases from diffs. Track coverage.

// memMemory layer

Persistent memory: PostgreSQL for structured data, Redis for session cache, Qdrant for vector search over embeddings.

// llmLocal LLM

Local models (Qwen, Llama) — no external APIs. Full autonomy, no code leakage to the cloud.

// mcpMCP servers

Model Context Protocol servers for integrating with external systems: Git, Docker, file ops, search.

// obsObservability

Prometheus + Grafana for metrics, structured JSON logs, distributed tracing via OpenTelemetry.