Skip to main content
AgentRuntime is an orchestration platform for production AI agents. You design multi-step workflows in the Console, connect third-party tools through MCP (Model Context Protocol), run workloads with retries and tracing, and govern access across tenants and projects. Think of it as Airflow for AI: event-driven workflow execution with native tool integration, not a generic chat UI.

​
What you can do

  • Design workflows in Workflow Studio — directed graphs of MCP calls, LLM steps, Lua scripts, loops, and human approvals
  • Connect tools via 40+ MCP connectors (Google Workspace, databases, Shopify, GitHub, and more)
  • Run reliably with dry-run validation, versioning, pause/resume, retries, and live event streams
  • Approve in the loop — human tasks surface in Command Center for review before a run continues
  • Observe usage — live runs, success rates, step traces, and credit consumption in Analytics
  • Automate via API — REST endpoints and personal access tokens for CI, scripts, and IDE integrations

​
Core components

ComponentWhat it does
ConsoleWeb UI at console.agentruntime.io — Workflow Studio, Command Center, Connections, MCP, Providers, Chat
BFF APIPublic API at api.agentruntime.io — auth, workflows, runs, billing, integrations
Workflow runtimeExecutes workflow graphs, manages parallelism, retries, and human-in-the-loop pauses
MCP layerCatalog of tool servers; tenant instances with config profiles and connection bindings
Agent serviceRuns built-in agent turns in chat and channel rooms
Platform MCPExposes Console APIs as MCP tools for IDEs at mcp.agentruntime.io

​
How it fits together

​
Key concepts

  • Workspace (tenant) — Your organization. Holds projects, members, billing, and integrations.
  • Project — Where workflows, runs, and MCP instances live. Roles are scoped per project.
  • Workflow — A versioned DAG of steps. Authored in Workflow Studio, validated before publish.
  • Run — One execution of a published workflow snapshot. Streams events over WebSocket.
  • Connection — Saved credentials (Google OAuth, WhatsApp, API keys) bound to MCP instances.
  • MCP instance — A deployed tool server in your project with configuration and connection overrides.
See Key concepts for the full model, or jump to the Quickstart to create your first workflow.

​
Who AgentRuntime is for

AgentRuntime is built for teams moving from AI experiments to production workloads:
  • Engineering teams automating ops, data, and integration tasks with agent workflows
  • Product teams that need human approval gates before agents take action
  • Organizations that require multi-tenant access control, usage metering, and audit trails
AgentRuntime is API-first. The Console is the primary operator interface, but every workflow capability is also available through the REST API and Platform MCP for programmatic use.