Skip to content

Architecture

TBD Agents is a distributed system designed to run custom AI agents at scale.

  • System Overview

    Components, request flow, and how the pieces connect.

    System Overview

  • Data Model

    Entity relationships and document schemas.

    Data Model

  • Scaling

    Horizontal scaling strategies for workers, API, and infrastructure.

    Scaling

System at a Glance

graph LR
    Client([Client / Dashboard]) -->|HTTP| API[FastAPI API]
    API -->|Enqueue| Redis[(Redis)]
    Redis -->|Task| Worker[Celery Workers]
    Worker -->|SDK Session| SDK[Copilot SDK]
    SDK --> Models[Copilot Models API]
    SDK --> MCP[MCP Servers]
    Worker -->|Publish Events| Redis
    Redis -->|Subscribe| API
    API -->|SSE Stream| Client
    Worker -->|Persist| Mongo[(MongoDB)]
    API -->|Read/Write| Mongo

Dashboard Integration Boundary

The Flutter dashboard is served from /dashboard, while the legacy static dashboard remains available at /dashboard-legacy. Its backend contract is already cleanly separated behind REST + streaming endpoints under /api.

  • CRUD resources: agents, MCP servers, custom tools, skills, knowledge sources/items, guardrails, tokens, providers, workflows, memories, and scheduled agents
  • Execution surfaces: POST /api/workflows/{id}/prompt, POST /api/workflows/{id}/halt, GET /api/workflows/{id}/stream, task history under /api/tasks
  • Chat surfaces: POST /api/agents/{id}/chat plus persisted session endpoints under /api/agents/{id}/chat/sessions
  • Dynamic form dependencies: model listing via /api/models, file uploads for knowledge items and custom tools, and workflow/task views that combine CRUD state with live streaming events

This makes the current UI a good candidate for a Flutter web replacement that preserves the feature set while keeping backend deployment and execution architecture unchanged.