Getting Started

Prerequisite
You need an LLM API key to use LLM nodes. Configure any provider on the Models page. Gemini, OpenAI, Anthropic, Groq, or a local Ollama instance. Function tools work without any key.

Quick start guide

1
Create your first agent
Go to Agents in the sidebar and click New Agent. This creates an empty agent and opens it directly in the Builder. Or use Imagine an Agent in the top bar to describe what you want and have it generated automatically.
2
Test an agent in the Builder
Click any agent to open it in the Builder. In the right panel under Test, type a message and hit Run. Tokens stream in real time; the Trace panel fills step by step.
Start with a simple one-node LLM agent. Then add a Tool node to see chaining.
3
Try Chat for a conversation
Open Chat in the sidebar. Select an agent on the left, then type. Responses stream in real time. If an agent hits a Human Review node, the pipeline pauses for approval from the Runs page.
4
Run from the API
Go to API Keys, create a key, then call your agent:
bash
curl -X POST https://your-domain.com/api/agents/AGENT_ID/run \
  -H "Content-Type: application/json" \
  -H "X-AgentHub-Key: ahk_your_api_key" \
  -d '{"message": "Hello!"}'

Imagine an Agent — AI-assisted building

Click Imagine an Agent in the top bar to describe what you want in plain English. The AI generates a complete agent plan with all nodes, connections, and configuration. You can then click Build to create it instantly or modify the plan before building.

Use this to scaffold complex agents (multi-step research, parallel analysis, HITL workflows) and then refine them in the Builder.
  • Describe what you want: "An agent that researches a company, drafts a report, and waits for human review before sending"
  • The AI generates a plan. Review it, then click Build.
  • The agent opens directly in the Builder ready to test.
  • Use the agent dropdown to modify an existing agent instead of creating a new one.

Sample agents reference

AgentNodesWhat it shows
Quick Chat
LLM
Simplest possible agent. One LLM node, no dependencies.
Text Analysis Pipeline
Action → Action → AI Step
Two function tools (no API key) feed results into an LLM.
Web Research Report
Action → Transform → AI Step
Search results formatted by Passthrough before LLM synthesises.
Email Drafter + Human Review
AI Step → Human Review → AI Step
LLM drafts, you review, LLM finalises.
Smart Support Router
AI Step → Branch → AI Step × 2
Classify input as BILLING or TECHNICAL, route to specialist.
Parallel Analyzer
Fork → [AI Step + Action] → Join → AI Step
Two branches run simultaneously, merged, then synthesised.
Iterative Blog Writer
Loop → AI Step (3×)
LLM drafts then improves the same post 3 times.
Topic Expert Router
Switch → AI Step × 4
LLM classification picks the right specialist branch.