Servflow - Build, Run, and Benchmark AI Agents

Agent designer and runtime.

Design your agent visually. Lay out its prompts, tools, decision branches and guardrails on a single canvas, then run it on the same platform you built it in.

Add a sub-agent, attach its tools, pick its model. See the whole architecture at a glance instead of tracing it through code.

Adding a sub-agent to an agent on the Servflow canvas

Visual canvasBranching logicActionModel-agnosticGit syncMCP

Built to handle complexity

Empowering technical and non-technical agent builders.

Product and ops teams design agents on the canvas. Engineers open the same agent as YAML to review it in a pull request, version it in Git, and ship it with the rest of the code.

No code

Drag in sub-agents, wire tools, pick models

Code

YAML config, Git history, one CLI command to create

name: Support triage
config:
  entries:
    - type: handler
      handler: github
      start: agent.classifier
  agents:
    classifier:
      providerID: 1
      systemPrompt: Label the issue.
      next: agent.researcher
    researcher:
      providerID: 2
      tools: [web_search, http, github_mcp]
      next: agent.responder
    responder:
      providerID: 1
      next: end

The Support triage agent on the visual canvas

Canvassupport-triage.yaml

Built to handle security

Built for security and production-grade load.

Agents run shell commands, read files and call tools. Servflow runs all of it inside an operating-system sandbox, on an engine written in Go so it stays small and quick.

Anthropic-grade sandboxing

Memory efficient, single Go binary

Fast, concurrent runs

sandbox · bwrap / seatbelt

shell · run 7f3a

$ git clone repo && make test

own folder · /runs/7f3a

Networkoff by default

~/.ssh, /etchost files hidden

Server processnot visible

one Go binaryconcurrent

run 1

run 2

run 3

run 4

run 5

secrets

{{ secret "github-token" }}

••••••••••resolved at run time

One-click runsActionLive test outputInstant re-runsAction

Demo

See it in action.

Watch an agent go from an empty canvas to a working run.

ToolRun action · MCP server · Call sub-agent · Agent Task Tool

Every action, MCP server and sub-agent is a tool.

Attach any of the 40+ built-in actions, point a sub-agent at an MCP server, or hand a task to another sub-agent on the canvas. Tools are wired visually and versioned with the agent.

HTTP RequestActionJavaScriptActionShell ScriptActionSend EmailActionDiscord Send MessageActionGitHub Fetch / PostActionMongoDB QueryActionStore / Fetch VectorsActionChromium NavigateActionJWT TokenActionGitHub MCPMCPAny MCP serverMCPCall sub-agentAgent Task Tool

Pricing

Choose your plan

Every plan is fully managed. Start free and upgrade as your agents take on more work.

Free

$0/ month

A permanent free tier for building and trying out agents.

Limits

10 agent runs per minute
3 enabled agents
3 concurrent runs

What's included

Fully managed hosting
Trace viewer

Get started

Starter

Pro

$50/ month

For agents running at higher volume and concurrency.

Limits

500 agent runs per minute
Unlimited enabled agents
50 concurrent runs

What's included

Fully managed hosting
Trace viewer

Choose Pro

Final response

Servflow on cloud, coming soon.

Sign up for early access, demos, and a chance to help guide product development.

Or join the community on Discord →

Servflow is not just another harness.