Foundry gives your LLM a corner office and a department budget

What it does

Foundry is an open-source platform that simulates a complete AI-powered company. You define a strategic goal, and a hierarchy of agents—Board, CEO, Department Heads, and Employees—breaks it down, delegates tasks, and reports back through a real-time group chat. A web-based admin panel lets you configure LLM keys, set budgets, and watch token costs accumulate in real time.

The interesting bit

The project treats multi-agent orchestration as an organizational design problem rather than a scripting challenge. It layers a three-tier memory system (company, department, agent) on top of PostgreSQL and pgvector, and insists on human-in-the-loop approval for critical decisions, which keeps the autonomous loop from running completely unsupervised.

Key highlights

• Auto-generates a corporate hierarchy (Board → CEO → Departments → Employees) from a name and industry template.
• Real-time group chat with streaming output, @mentions, and human approval gates for critical decisions.
• Built-in cost governance: token tracking, budget caps, model routing, and an LLM key pool for rotation.
• Three-tier memory with RAG retrieval and automatic consolidation, backed by PostgreSQL/pgvector.
• Full-stack TypeScript monorepo: NestJS backend services, React frontend, Docker Compose for infrastructure.

Caveats

• The README warns of ~2 GB of Docker image downloads and an 8 GB RAM minimum, with first launch taking 5–10 minutes.
• AI agents will not respond until you manually configure LLM API keys (OpenAI, Claude, etc.) in the admin panel; there is no built-in model.
• The architecture diagram and service list suggest significant operational complexity—PostgreSQL, Redis, RabbitMQ, Consul, and multiple NestJS services are all required to run the “company.”

Verdict

Foundry is worth a look if you want to experiment with autonomous multi-agent workflows wrapped in a believable corporate interface. If you just need a single LLM call or a lightweight agent script, this is organizational overkill.