Platform Overview
SixDegree is a unified platform for discovering, mapping, and acting on your organization through AI-powered interaction with your knowledge graph.
Architecture
Core components
Dashboard
The web interface at app.sixdegree.ai provides visual access to your knowledge graph. Navigate environments, explore the entity ontology, chat with AI, configure molecules, manage agents and triggers, and review the approval queue.
CLI
The degree command-line tool enables terminal-based interaction. Use it for interactive chat, entity queries, automation scripts, and CI/CD integration. Install with curl -fsSL https://get.sixdegree.ai | sh or via Homebrew.
Platform API
REST API for programmatic access to all platform features. OAuth2 authentication with full SDK support in Go and TypeScript. Major surfaces include entities, discovery, chat completions, agent tasks, triggers, approvals, memory, invitations, and webhook ingest.
Molecules
Extensible plugins that discover entities from external services and provide AI tools for taking actions. Each molecule connects to services like Salesforce, Zendesk, Stripe, NetSuite, GitHub, or custom systems. Molecules can provide any combination of discovery, MCP tools, skills, visualizations, and shipped agents that ride alongside the integration. Learn more in Molecules.
Knowledge graph
Entities and relationships are stored as a graph database. Real-time queries power AI chat, the visual ontology, and the late-disclosure mechanism that decides which molecule tools to expose to agents. The graph scales to millions of entities per environment with optimized traversal queries.
Model Context Protocol
MCP tools give AI the ability to take actions in your services. Molecules expose tools like logging a call on an account, replying to a support ticket, pulling invoices, or triggering workflows. The platform also ships built-in tools (plan, memory, ontology, aggregate, calc) available to every agent regardless of which molecules are configured.
Agents, triggers, and approvals
Named, reusable AI specifications that run on your graph. Agents carry their own autonomy tier, scope, and blast-radius limits. Triggers fire them on events, schedules, or webhooks. Approvals gate writes and destructive actions, with MFA step-up required for destructive approvals.
Task queue
Agent runs are durable tasks persisted in sixdegree.agent_tasks and dispatched through a River-backed queue. Tasks survive process restarts, can be cancelled mid-flight, and graph-shaped runs (workflows) gate dependent nodes atomically.
How it works
Discovery. When you configure a molecule, it discovers entities from external services and maps relationships between them. The platform schedules regular discovery runs to keep your graph current, or receives webhook updates for real-time changes.
Chat. When you chat with AI, it queries the knowledge graph to understand your organization and selects relevant MCP tools to take actions. Late tool disclosure ensures the AI only sees tools relevant to the entities in play. No Intercom tools surface when you're working with Zendesk entities.
Agents. When you save an agent, it becomes a named, reusable spec that triggers can fire and humans can invoke. Each invocation is a durable task with a plan, a run trace, and a final result. Writes go through the approval queue; destructive actions require MFA step-up.
Infrastructure as code. The Terraform provider manages environments, molecules, agents, and triggers as code.
Deployment
SaaS
Hosted at app.sixdegree.ai with automatic updates, built-in high availability, and zero infrastructure to manage. Choose SaaS for faster setup and managed operations.
Self-Hosted
Deploy in your own Kubernetes cluster when you need air-gapped environments or strict data residency requirements. Contact sales for deployment guides and infrastructure requirements.
Next steps
- Entities & Relationships: understanding your graph
- Molecules: discovery and extensibility
- Agents: named, reusable AI specifications
- Triggers: fire agents on events, schedules, or webhooks
- Approvals: human-in-the-loop for writes