Guide

AI Agent API vs LLM API: Key Differences

The Core Difference

LLM API (OpenAI, Anthropic, Google): You send messages, get back text. You write the prompt, handle the output, parse results, and manage all the application logic.

AI Agent API (Runcept): You send typed input, get back structured JSON. The prompt, parsing, retries, and output formatting are baked into the agent.

Side-by-Side

LLM API call:
POST /v1/messages
{ "model": "claude-sonnet-4-6", "messages": [{"role": "user", "content": "Review this code: ..."}] }
→ { "content": [{"type": "text", "text": "Here are my thoughts on the code..."}] }

Agent API call:
POST /api/v1/run
{ "agent": "code-reviewer", "input": { "diff": "..." } }
→ { "output": { "score": 82, "issues": [...], "suggestions": [...] } }

When to Use an LLM API

You're building a custom agent with specific behavior

You need fine-grained control over the prompt

You're experimenting with different models or prompt strategies

The task is unique enough that no pre-built agent exists

When to Use an Agent API

You want to complete a well-defined task (code review, PR description, summarization)

You don't want to write or maintain prompts

You need structured JSON output you can use programmatically

You want predictable pricing per task rather than per token

You're integrating into a CI pipeline or automation script

The Build vs Buy Decision

Do you need custom behavior? Yes → use an LLM API, build your own No → is there an agent on Runcept that does this? Yes → call it via Agent API No → build it, publish it, earn per run

Combining Both

Many builders use LLM APIs internally to power their agents. You call the Agent API to consume the task; the agent uses an LLM API internally to do the work. These layers aren't competing — they compose.

Next Steps

Browse agents on Runcept for your task

Call your first agent

Build and publish an agent if you need custom behavior