Cognition Documentation¶
Cognition is a headless agent orchestration backend. Define your agent and get REST API, streaming, persistence, sandboxing, multi-user scoping, and observability automatically.
This site is organized around four documentation surfaces:
- Architecture records code-derived C4 views, runtime flows, operational constraints, and durable decisions.
- Concepts explain individual subsystems.
- Guides show how to perform concrete tasks.
- Blueprints describe reference architectures for products built on Cognition.
If you want a course sequence rather than reference material, use Learn agent development.
Start Here¶
- Getting Started: install, configure, and send your first message.
- Core vs App Layer: understand what Cognition owns versus what your product owns.
- Architecture: explore code-derived C4 diagrams, component boundaries, runtime flows, deployment topology, and decisions.
- Extending Agents: add memory, skills, tools, subagents, middleware, MCP servers, and custom providers.
- A2A Builder Guide: publish, describe, and securely expose agents through A2A 1.0.
- AWS Lambda MicroVM Sandbox Setup: configure AWS-native MicroVM sandbox isolation for agent commands.
Concepts¶
| Document | Description |
|---|---|
| Architecture | Code-derived C4 views, runtime flows, deployment topology, risks, and decisions |
| Sessions & Messages | Session lifecycle, message persistence, SSE streaming, and reconnection |
| Agent Runtime | Agent runtime protocol, definitions, A2A exposure, and multi-agent registry |
| A2A in Cognition | Agent Cards, message Parts, tasks, streaming, artifacts, security, and scoping |
| Storage & Execution | Storage and execution backend protocols and implementations |
| Sandboxes | Local, Docker, Kubernetes, and AWS Lambda MicroVM execution backends |
| Kubernetes Sandbox | K8s-native sandbox isolation using the agent-sandbox CRD and controller |
| AWS Lambda MicroVM Sandbox | AWS-native MicroVM sandbox isolation, profiles, IAM roles, networking, and runtime protocol |
| Observability | OpenTelemetry traces, Prometheus metrics, and MLflow experiment tracking |
| Security | Scoping, sandbox isolation, tool security, MCP policy, A2A boundaries, rate limiting, and CORS |
Guides¶
| Document | Description |
|---|---|
| Getting Started | Install, configure, and send your first message |
| Core vs App Layer | Builder responsibilities: what belongs in Cognition versus your product |
| Configuration | YAML keys and environment variables |
| Extending Agents | Add memory, skills, tools, subagents, middleware, MCP servers, A2A exposure, and custom providers |
| A2A Builder Guide | Configure Agent Cards, public skills, media modes, message Parts, authentication discovery, and scope isolation |
| Deployment | Docker Compose, PostgreSQL, migrations, and production hardening |
| API Reference | REST endpoints, SSE events, artifacts, A2A protocol, capabilities, and scoping headers |
| Release Checklist | Release process for Cognition versions |
Blueprints¶
| Blueprint | Domain |
|---|---|
| Cognition CLI | Terminal assistant over the Cognition API |
| BreachLens | Cybersecurity SOC investigation and triage |
| GeneSmith | Secure biological foundry for protein design |
| DataLens | Headless data science on sensitive datasets |
| StarKeep | SpaceOps administrator for satellite repair |