Explanation
Background, architecture, and design rationale. This section answers why things work the way they do — read it when you're curious about the thinking behind the platform.
These aren't step-by-step guides. They're discussions that help you build a mental model of the system so you can make better decisions when building on it. These explanations are written for platform users and developers who want to understand the design decisions behind the platform.
Architecture & Core Systems
| Topic | What it covers |
|---|---|
| Platform Architecture | Tech stack, system layers, and how a request flows through the platform |
| Agent Messaging System | RabbitMQ-based message dispatch, tool-call loops, and deduplication |
| Task Execution Lifecycle | The three state machines that drive agent work — tasks, idle, and conversations |
| Conversation Model | How conversations, participants, and message dispatch work |
| Org, Workspace & Project Hierarchy | Multi-tenancy model and why "Workspace" was renamed to "Desktop" |
| Agent Deployment Model | Knative services, NFS mounts, and the trusted/untrusted cluster split |
Features Deep-Dive
| Topic | What it covers |
|---|---|
| Knowledge & Memory | Short-term, mid-term (Vecto), and long-term memory tiers |
| Security Model | OIDC hybrid auth, token validation, file sandboxing |
| Billing & Subscriptions | Stripe integration, Keycloak attributes, tier enforcement |
| Desktop & Computer Use | VMs, VNC, robot API, and the computer-use automation loop |
| Email System Design | SendGrid gateway, routing, rules, and approval workflows |
| Xircuits & Visual Programming | What Xircuits is and how it fits into the agent ecosystem |
| Voice Call Architecture | LiveKit, OpenAI Realtime API, and turn-taking protocol |
| Integration Patterns | How workspace-level integrations flow into agent pods |