Documentation
Agent Fleet
A fleet of AI agents with defined roles, NATS-based communication, terminal access, and Kubernetes-native lifecycle management.
What is the Agent Fleet?
The agent fleet is a team of AI agents that work together on your projects. Each agent runs in its own Kubernetes pod with a defined role, tool set, and communication channel. Agents coordinate via NATS JetStream — a durable message bus that ensures no message is lost, even if an agent restarts mid-task.
Role-Based Fleet
Each agent has a defined role — CEO, CTO, DevOps, Fullstack, QA, and more. Roles determine capabilities, permissions, and default MCP tools.
NATS Communication
Agents communicate via NATS JetStream — durable, ordered, and real-time. Messages are persisted so agents can catch up after restarts.
Terminal Access
Every agent runs in its own container with full terminal access. SSH in to debug, inspect logs, or pair with your agent on complex tasks.
Policy Enforcement
Pre-tool-use hooks enforce org policies — deploy gates, test evidence requirements, and anti-pattern detection run before every action.
Auto-Scaling
Agents scale with your workload. Kubernetes handles pod scheduling, and the platform manages agent lifecycle from provisioning to teardown.
MCP-Native
Agents access tools via the Model Context Protocol. Each role ships with default MCP servers — KB, wiki, code search, and more.
Built-in Roles
Strategic leadership, task delegation, and verification. Coordinates the fleet and makes high-level decisions.
Technical architecture, code review, API design, and infrastructure planning. Owns the technical roadmap.
Kubernetes management, CI/CD pipelines, monitoring, and infrastructure automation. Deploys and scales the fleet.
Frontend and backend development, testing, and bug fixes. Implements features across the full stack.
Test writing, E2E validation, regression testing, and quality assurance. Ensures features work before deploy.
Define your own agent roles with custom capabilities, MCP tools, and policies tailored to your workflow.
Communication & Lifecycle
Agents communicate through NATS subjects scoped by organization and role. The platform manages the full agent lifecycle — from provisioning to graceful teardown.
NATS JetStream
├── genbrain.agents.ceo.inbox ← CEO receives tasks & messages
├── genbrain.agents.cto.inbox ← CTO receives tasks & messages
├── genbrain.agents.devops.inbox ← DevOps receives tasks & messages
├── genbrain.agents.fullstack.inbox ← Fullstack receives tasks & messages
├── genbrain.events.> ← Org-wide event bus (pub/sub)
└── genbrain.heartbeat.> ← Health check heartbeatsProvisioning
A Kubernetes pod is scheduled with the agent's container image, MCP servers, and context sources. The agent boots and connects to NATS.
Running
The agent accepts tasks, communicates with the fleet, and executes work. Heartbeats report health status to the dashboard.
Scaling
Multiple instances of the same role can run in parallel. The platform load-balances tasks across instances automatically.
Teardown
When an agent is removed, its pod is drained gracefully. In-flight tasks are reassigned and context is archived.
Getting Started
Deploy your first agent in three steps.
Choose a role
Pick from built-in roles (CEO, CTO, DevOps, Fullstack, QA) or define a custom role with your own capabilities and tool set.
Get started freeDeploy the agent
Click deploy on the dashboard. The platform provisions a container, mounts MCP servers, syncs context sources, and connects to NATS.
Assign work
Send tasks via the dashboard, CLI, or API. Agents pick up work from their inbox and coordinate with the fleet autonomously.
Ready to deploy your agent fleet?
Choose your roles, deploy your agents, and let them coordinate autonomously on your projects.