Skip to main content
Agents

Agents

Agents are autonomous AI workflows made of composable steps, skills, and tools. Build multi-step pipelines that plan, research, execute, and review — then call them with a single API request via the OpenAI-compatible endpoint.

One API call, multi-step execution

Set model: "agent:your-agent-id" in any OpenAI-compatible request. Agentlify orchestrates every step, tool call, and skill lookup server-side, then returns a single response — or streams it token-by-token.

Quick Example

javascript
// Agents use a dedicated endpoint: /api/agents
const response = await fetch('https://agentlify.co/api/agents', {
  method: 'POST',
  headers: {
    'Authorization': `Bearer ${process.env.AGENTLIFY_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: 'my-research-agent',   // agent ID = the slug
    messages: [{ role: 'user', content: 'Compare React and Vue in 2025' }],
    stream: true,
  }),
});

const reader = response.body.getReader();
const decoder = new TextDecoder();
while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  process.stdout.write(decoder.decode(value));
}

How Agents Work

Every agent is a pipeline of steps executed in order. Each step can call an LLM with a prompt template, invoke a built-in tool, or trigger a webhook. Steps pass context forward via variables like {{previous_output}} and {{input}}.

1

User sends a message

Via API (model: "agent:slug") or the in-app test panel.

2

Orchestrator runs each step

LLM prompts are sent to your router. Tool steps execute server-side. Each step's output feeds into the next.

3

Final response returned

Standard OpenAI completion format with an extra _meta field containing cost, latency, and step details.

Core Concepts

Steps & Prompts
The building blocks. Each step is an LLM call with a prompt template. Chain them together to create complex workflows.
Skills
Instructional knowledge packs the agent can retrieve on demand. Add domain expertise without bloating the system prompt.
Built-in Tools
Calculator, Web Search, Code Executor, JSON Formatter, and Text Analyzer — ready to use, no setup required.
Custom Tools
Define your own tools with webhook endpoints. The agent calls your API when it needs external data or actions.
Settings
Planning mode, timeouts, max steps, streaming behavior, and tool execution backends.
Agent API Reference
Full API docs: request format, response schema, streaming, error codes, and SDK examples.

Anatomy of an Agent

json
{
  "model": "my-research-agent",       // Slug — also the Firestore doc ID
  "displayName": "Research Agent",       // Human-readable name (editable)
  "description": "Researches topics and produces structured summaries",
  "isActive": true,

  "steps": [                             // Executed in order
    {
      "id": "step-1",
      "name": "research",
      "displayName": "Research Phase",
      "runtime": "llm",                  // "llm" or "tool"
      "phase": "research",               // UI label: research | plan | execute | review
      "prompt": "Research the following topic: {{input}}",
      "order": 0,
      "enabled": true
    },
    {
      "id": "step-2",
      "name": "summarize",
      "displayName": "Summarize",
      "runtime": "llm",
      "phase": "execute",
      "prompt": "Summarize: {{previous_output}}",
      "order": 1,
      "enabled": true
    }
  ],

  "skills": ["code-review-guidelines"],  // Instructional skills (via get_skill)
  "builtinTools": ["builtin_web_search"], // Executable tools
  "tools": [],                            // Custom webhook tools

  "settings": {
    "planningMode": "initial_plan",       // off | initial_plan | per_step_plan
    "timeout": 120,                       // seconds
    "streamFinalOnly": true,
    "defaultToolBackend": "client"        // client | webhook
  }
}

Next Steps

Create your first agent

Step-by-step walkthrough from template to production.

Call an agent via API

Use the OpenAI SDK with model: "agent:slug".