Skip to main content
Forge is already deployed and running. This guide walks you through the platform’s core surfaces in a logical sequence — architecture, agents, workflows, governance, and the API. Each step links to the detailed documentation for that area.
This is an orientation guide, not a setup tutorial. Forge is a managed platform — there is nothing to install.
1

🏗️ Understand the architecture

Forge operates across four layers: (durable backend), (operator dashboard), (execution router), and (pluggable runtimes). Understanding this topology is the foundation for everything else.
ConvexConvex

Source of truth

117 tablesInterpreter
BurgundyBurgundy

Control plane

85+ routesApprovals
BridgeBridge

Execution router

AdaptersSSE
🔌 Gateways

Runtimes

OpenClawClaude

Platform Overview

Full architecture diagram, data flow, and capability map.
2

🤖 Explore agents

Agents are the actors in Forge. They have definitions (what they are), deployments (where they run), skills (what they know), and tools (what they can do). The agent registry in Convex is the source of truth for declared state; gateways hold runtime state.Key agent surfaces:
  • Agent definitions — identity, model config, tool profiles, workspace files
  • Agent templates — pre-built configurations for one-click deployment
  • Skills and tools — the capability layers that determine what an agent can do
  • Sessions — bounded interaction contexts with observable traces
  • Memory — 3-layer model (session, channel, long-term) for context persistence

Agents Overview

Agent architecture, registries, and the capability model.
3

⚙️ Run a workflow

Workflows are defined in — a YAML DSL that declares steps, dependencies, roles, and failure handlers. The Convex interpreter parses definitions into a and drives execution through dispatch, governance, and state advancement.The execution lifecycle:
  1. .lobsterX definition is parsed and validated
  2. DAG is resolved from step dependencies
  3. Each step passes through the 10-gate governance pipeline
  4. Allowed steps are dispatched to gateways via Bridge
  5. Results flow back; the interpreter advances the DAG
  6. Checkpoints enable rewind and branch at any point

Workflows Overview

.lobsterX format, DAG execution, and checkpoints.
4

🛡️ Set up governance

Every step dispatch passes through a 10-gate safety pipeline before execution. Gates check gateway health, agent availability, concurrency, rate limits, budgets, trust levels, context trust, policies, and approval status. The pipeline produces a with a disposition of allow, block, or hold.

10-Gate Safety Pipeline

Gateway HealthAgent StatusConcurrencyRate LimitBudget (Agent)Budget (Envelopes)Trust LevelContext TrustPolicy RulesApproval

Governance Overview

Safety gates, policies, approvals, and budgets.
5

🔌 Connect via API

The Platform API exposes 42+ REST endpoints under an OpenAPI 3.1 spec. Manage agents, trigger workflows, query run status, retrieve traces, and configure governance programmatically. Authentication uses API keys managed through the Burgundy dashboard.Available integration surfaces:
  • REST API — full CRUD for agents, workflows, governance, and observability
  • TypeScript SDK — typed client library wrapping all API operations
  • MCP Server — Model Context Protocol server for agent-to-Forge integration
  • Webhooks — configurable event notifications for workflow state changes

API Overview

Endpoints, authentication, and the OpenAPI spec.

​
🚀 Next Steps

📊 Observability

Traces, analytics, and audit trails.

🏭 Factories

Autonomous production units with agent teams.

📦 TypeScript SDK

Typed client library and MCP server.