AgentHub is a visual AI agent builder. You connect nodes on a canvas — AI Step (LLM), Action (tools), Branch, Switch, Loop, Fork, Join, Human Review, Ask User — and every agent is automatically deployed as a REST API.

What LLM providers does AgentHub support?

AgentHub supports Google Gemini, OpenAI, Anthropic Claude, Groq, Ollama, and any OpenAI-compatible API endpoint including LM Studio and Mistral.

How do I call an AgentHub agent from my application?

Every agent is available as a REST API. Send a POST request to /api/agents/{agentId}/run with your message and X-AgentHub-Key header. The response includes the output, token count, and execution trace.

What is a Human Review node?

A Human Review (HITL) node pauses the agent pipeline and waits for a human to approve, request revision, or reject before the agent continues. Useful for content approval, compliance checks, or any irreversible action.

Can agents run in parallel?

Yes. Fork nodes split execution into multiple parallel branches that all run simultaneously. A Join node collects the results and merges them as an array, object, or concatenated text.

AgentHub — Visual AI Agent Builder

Build multi-step AI agent pipelines on a visual drag-and-drop canvas. Connect AI Step nodes (LLMs), Action nodes (tools, web search, code execution), Branch and Switch nodes for routing, Loop nodes for iteration, Fork and Join nodes for parallel execution, Human Review nodes for human-in-the-loop approval, and Ask User nodes for mid-run clarification. Deploy every agent as a live REST API with one click.

Node Types

AI Step: call any language model (OpenAI GPT-4o, Google Gemini, Anthropic Claude, Groq Llama, Ollama) with a system prompt and conversation memory.
Action: run tools including HTTP API calls, web search (DuckDuckGo, Tavily, Serper), web scraping (Jina Reader), Python/JavaScript/Bash code execution, and datatable operations.
Branch: binary yes/no conditional routing evaluated by an AI model.
Switch: multi-way routing for 3 or more branches using LLM classification.
Loop: repeat a section of the pipeline until an exit condition is met or a maximum iteration count is reached.
Fork and Join: split execution into parallel branches that run simultaneously, then merge results.
Human Review: pause the pipeline for human approval, revision request, or rejection before continuing.
Ask User: pause and ask the user a clarifying question, then continue with their answer.
Transform: pass data through with optional template transformations.
Start and End: required entry and exit points for every agent pipeline.

Key Features

Visual canvas: drag, drop, and connect nodes to build AI pipelines.
REST API deployment: every agent is immediately available as a POST endpoint.
Real-time streaming: SSE streaming with token-by-token output.
Execution trace: waterfall chart and step-by-step trace for every run.
Memory: sliding window, full history, or AI-summarized conversation memory.
Guardrails: input and output safety rules with keyword and pattern matching.
Prompts library: reusable system prompts with template variable support.
Datatables: structured data tables for import (LLM context) and export (structured output).
Orchestrator: smart clarify routing that classifies user messages during agent pauses.
Multi-provider LLM: OpenAI, Google Gemini, Anthropic, Groq, Ollama, and any OpenAI-compatible endpoint.

Visual AI Agent Builder · Bring Your Own API Keys

Build AI agents
visually. Deploy as
REST APIs.

Production-grade AI agent infrastructure. Orchestrate LLMs, tools, and MCP servers with parallel execution, conditional routing, memory, and human oversight. Every agent ships as a REST API.

Build via Chat · now live

Multi-agent systems · built in

View Docs Contact Us

Free to start

No credit card

Your own API keys

No vendor lock-in

research-agent.flowStep 2/4 · tool running Executing

POST /api/agents/abc123/run stream: true

{ "status": "running", "step": 2, "tokens": 127 }

Works with any LLM. Bring your own API keys.

OpenAIGeminiClaudeGroqMistralLlamaOllamaAny OpenAI-compatible APIOpenAIGeminiClaudeGroqMistralLlamaOllamaAny OpenAI-compatible APIOpenAIGeminiClaudeGroqMistralLlamaOllamaAny OpenAI-compatible API

Everything you need

Every building block
for serious AI systems.

AgentHub gives you the primitives to build production-grade AI workflows without writing orchestration logic from scratch.

LLM AgentsVisual BuilderREST APIsLoops + ParallelHuman-in-the-LoopGuardrailsMCP ConnectorsFull Observability

Core

Visual DAG Builder

Build complex multi-step AI pipelines by connecting nodes on a drag-and-drop canvas. Each node does exactly one thing, making agents easy to debug, extend, and understand.

Instant REST API

Every agent auto-gets a live POST endpoint. No deployment steps. Call from anywhere in seconds.

Any LLM Provider

Bring your own keys for OpenAI, Gemini, Claude, Groq, Mistral, or any compatible endpoint.

Conditional Routing

Branch and Switch nodes route based on what the model finds at runtime.

Loops + Parallel

Loop until quality passes. Fork/Join runs branches simultaneously.

Human-in-the-Loop

Human Review nodes pause for human approval before the agent continues.

Guardrails

Block bad input and filter unsafe output per-node, not globally.

Build here. Integrate anywhere.

Build your agent once.
Call it from any system.

Every agent you build instantly becomes a REST endpoint. Drop it into any product, backend service, or automation pipeline. JavaScript, Python, cURL, or any HTTP client. One POST call is all it takes.

// JavaScript (fetch)
const res = await fetch("https://app.com/api/agents/ID/run", {
  method:"POST",
  headers:{ "X-AgentHub-Key":"ahk_xxxx" },
  body: JSON.stringify({ message:"Summarise this..." })
});
const data = await res.json();
// { status:"completed", tokens:342 }

How it works

Three steps from idea
to running API.

Design on the canvas

Drag nodes onto the canvas and connect them. Each node does one thing: call an LLM, run a tool, check a condition, or wait for approval.

Pick any LLM

Configure your own API keys for OpenAI, Gemini, Claude, Groq, Mistral, or any OpenAI-compatible endpoint including self-hosted Ollama.

Call it instantly

Every agent is immediately available as a REST API. One POST request and your agent runs. Tools fire, LLMs respond, results return.

Get in touch

Have a question or
feature request?

Drop us a line. We read every message and respond within 24 hours.

Tokens used

1,204

this run

Latency

1.2s

end-to-end

Cost estimate

$0.0024

per run

Nodes executed

4/4

all passed

Full observability

Every run is logged.
Nothing is a black box.

Token usage, latency, cost attribution, full execution trace, and node-level outputs, all stored and searchable.

Per-node token and latency breakdown

Cost per run with model attribution

Full execution trace with timestamps

Filter, search, and export run history

Why AgentHub exists

Most AI tooling today
is shallow.

We built AgentHub because production AI systems need real infrastructure, not drag-and-drop prompt chains dressed up as agents.

Already here

Multi-agent Systems · built in

Every pipeline is a multi-agent system. Each node runs with its own LLM, tools, and memory. Connect any topology on the same visual canvas.

Agents should be composable

Every complex workflow breaks down into simple, testable nodes. Build small, compose big. Swap out any part without breaking the whole.

Bring your own everything

Your API keys, your models, your tools. No lock-in. No hidden costs. No surprises. AgentHub is infrastructure, not a platform moat.

Observability is not optional

Production agents need full trace logs, token counts, cost attribution, and error reporting. Not as an afterthought. By default.

Humans stay in control

The best AI systems know when to pause and ask. HITL checkpoints and approval gates make that a first-class primitive, not an edge case.