Compare · camelRun by camelAI

Managed agents compared

camelRun by camelAI runs your AI agent's loop on the model you pick, hosted by us or on your own servers.

Where it runs

PlatformcamelRunHill: it runs the agent loop for youSmall hill: for some agent types

Managed runtime

camelRun

camelAI

Models from many makersHosted or on your own servers

Models: Claude, GPT, Gemini, open models, or a model server of your own
Runs on: run.camelai.com, or your own servers with one container and Postgres.
Agent loop: camelRun runs it. Your tools stay in your own code.

How camelRun works

PlatformcamelRunHill: it runs the agent loop for youSmall hill: for some agent types

The field

How managed agent platforms differ

Four kinds of platform run AI agents today. camelRun runs the loop for you, as the labs' managed agents do, on models from many makers, hosted or on your own servers.

Managed agents from a model lab

Pick one when your team has settled on one lab's models and wants the agent hosted next to them.

Claude Managed AgentsClaude 4.5 and later · Anthropic's platform
OpenAI Agents APIOpenAI's models · OpenAI's platform
Managed Agents in the Gemini APIGemini, through Antigravity · Google's cloud
Amazon Bedrock Managed AgentsBuilt for OpenAI's models · AWS

Cloud agent platforms

Pick one when your company already builds on that cloud and wants its agents there too.

Amazon Bedrock AgentCoreAny model · AWS
Google Agent RuntimeAlmost any model · Google Cloud
Microsoft Foundry Agent ServiceModels from the Foundry catalog · Azure

Agent frameworks with their own hosting

Pick one when you've built on that framework and want hosting made for it.

Durable execution engines

Pick one when you want to write every step of the loop yourself, as a workflow.

TemporalAny model · Temporal Cloud or your own servers

camelRun up close

Managed AI agents on your model

run 1 · openrouter/openai/gpt-6-luna
run 2 · openai/gpt-5.2
run 3 · amazon-bedrock/global.anthropic.claude-sonnet-5-5
run 4 · anthropic/claude-sonnet-5-5
run 5 · home-ollama/qwen3:32b

Inset A · Agenthello-ts

Hosted: new Agents()
Your servers: new Agents({ url, apiKey })

camelRun runs the loop and keeps each agent's history and files. Change an agent's model between runs with agent.configure({ model }) and its history carries over. Your tools stay in your own code, with your keys.

camelAI's own Discord bot builder, camelBot, runs on camelRun.

index.ts

1import { Agents } from "@camelai/run";
2
3const agents = new Agents(); // reads CAMELAI_API_KEY
4const agent = await agents.upsert("hello-ts", { model: "openrouter/openai/gpt-6-luna" });
5console.log((await agent.run("Say hello in three languages.")).text);
6await agents.close();

Questions

Questions about managed agents

What are managed agents?

A managed agent is an AI agent whose loop a platform runs for you: it calls the model, runs the tools and keeps the history. Claude Managed Agents, OpenAI's Agents API and Google's Managed Agents do this on their own models. camelRun does it on models from many makers.

How much does it cost to use a Claude Managed agent?

Claude Managed Agents bills tokens at Claude's API rates plus $0.08 per session-hour while a session is running (Anthropic's pricing page, checked October 23, 2026). On camelRun, an agent costs $0.01 per agent-hour while it's in a run and $0.10 per GB-month of storage. Idle and sleeping agents cost only storage, and tokens on your own provider keys cost $0.

How are Claude, OpenAI and Google managed agents different?

Each one runs its maker's models. Claude Managed Agents runs Claude 4.5 and later. OpenAI's Agents API runs OpenAI's models. Google's Managed Agents, in preview in the Gemini API, runs Gemini through Google's Antigravity agent. camelRun runs your agent on models from all three makers and more, and you can switch between them.

What's the difference between a managed agent and an agent framework?

With a framework such as LangGraph or the OpenAI Agents SDK, you build the agent in your code and host it yourself, or on a host like LangSmith Deployment. A managed agent platform runs the loop for you and keeps the history. camelRun is managed, and your tools still run in your own code.

Can I run managed agents on my own servers?

Yes. camelRun's runtime is open source under the AGPL and runs anywhere with Docker: one container and Postgres, with the same SDK and API as run.camelai.com. Letta and Temporal run on your servers too, and LangSmith Deployment does on its Enterprise plan. Anthropic and OpenAI let you host their agents' sandbox while they run the loop.

Can I change my agent's model later?

Yes. Change the model between runs and the agent keeps its history. Use models from Anthropic, OpenAI, Google, OpenRouter, Amazon Bedrock and more, on our keys or your own, or add a model server of your own that speaks OpenAI's or Anthropic's API.

Which managed agent platform should I choose?

Start from your situation. If your team has settled on one lab's models, that lab's managed agents keep everything in one place. If you want models from many makers, hosted or on your own servers, camelRun runs the loop for you, and the console's Export data gives you every agent's whole history as one zip whenever you want it.

Directions

Start your first managed agent on camelRun

Get an API key

$ export CAMELAI_API_KEY=art_...

By hand

Install the SDK

TypeScript or Python, in the app you already run.

$ npm install @camelai/run

$ pip install camelai-run

Run your first agent

The quickstart is one file: a tool, an agent and its first run.

Read the quickstart

Or skip the steps

Set up with your coding agent

Read https://run.camelai.com/SKILL.md and set up camelRun in this project. Ask me what the agent should do.

Paste into Claude Code, Codex, Cursor or any coding agent. It asks what your agent should do, then builds it.

Your agent runs on camelRun

hello-ts · run 1 · completed