CAMELRUN · OPEN SOURCE · AGPL

Cloud hosting for AI agents

Define an agent's model, instructions, tools, and channels. camelRun runs the agent loop, keeps its history, and wakes it when there is work.

Get an API key Read the quickstart
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.

agent.ts
import { Agents } from "@camelai/run";

const agents = new Agents(); // CAMELAI_API_KEY
const agent = await agents.upsert("support-triage", { instructions: "Be concise." });
console.log((await agent.run("Which tickets look urgent?")).text);
await agents.close();

What it is

Deploy AI agents without managing infrastructure

Most teams putting an AI agent in their product build the agent loop and the hosting themselves. camelRun is the off-the-shelf version, the way teams use Postgres instead of writing their own database.

Read why we built it
You define
Model
anthropic/claude-sonnet-5-5
Instructions
"Be concise."
Tools
weather
Channels
slack
camelRun runs
  • The agent
    loop
  • History and compaction
  • Code sandbox
  • Wake-ups and schedules
  • Scaling to thousands of agents

Scale

Run thousands of AI agents at once

Give every customer, ticket, or pull request its own agent, and scale up without talking to sales.

Durability

AI agents that survive deploys and crashes

An agent's history, files, and pending work live in camelRun, not in your app. When a server goes away, another one picks up where it left off.

Tools

Connect your tools with OpenAPI, MCP, or your own code

camelRun makes every tool call, so the model never sees your credentials. If your app stops answering, the call fails and the agent keeps working.

OpenAPI spec

Every operation becomes a tool.

MCP server

Remote servers, including ones that need auth.

Your code

TypeScript or Python functions, no public endpoint.

Built-in tools

Web search, web fetch, schedules, and asking the user.

Channels

Put your agent in Slack, Discord, email, GitHub, or your own app

Connect a channel in the console and every thread, email, or pull request gets its own agent. Telegram and webhooks from services like Sentry, Linear, and Stripe work too.

One agent per thread.

alice: @agent Delete the preview environment pr-4312.

agent: Approve this action? delete_environment (application) with {"name":"pr-4312"} Reply approve or deny.

alice: approve

agent: Deleted pr-4312.

No lock-in

Use any model, bring your own keys, or host camelRun yourself

Switch an agent's model at any time and it keeps its history. camelRun is open source under the AGPL. Self-hosting takes one Docker container and Postgres.

View source on GitHub
  • Anthropic, model id anthropic/claude-sonnet-5-5
  • OpenAI, model id openai/gpt-5.2
  • OpenRouter, model id openrouter/anthropic/claude-sonnet-5
  • Ollama, model id home-ollama/qwen3:32b
  • Amazon Bedrock, model id amazon-bedrock/global.anthropic.claude-sonnet-5-5
  • vLLM

Pricing

Pay only for the time your agents spend working

Idle and sleeping agents cost only storage.

$0.01 per agent-hour of active run time

$5
Free credit

For GitHub accounts older than 30 days. Everyone else verifies a card, with no charge.

Get an API key
Storage
$0.10per GB-month
Model tokens
$0 on your keysList price on our keys, no markup
Web search
$0.001 to $0.007per search

FAQ

Questions about camelRun

camelRun is a hosted runtime for AI agents from camelAI. You define an agent's model, instructions, tools, and channels. camelRun runs the agent loop, stores its history and files, runs model-written code in a sandbox, and scales your agents.

Agent time costs $0.01 per agent-hour while an agent is in a run. Storage costs $0.10 per GB-month. Tokens are free on your own provider keys and cost the provider's list price on ours.

Billing is prepaid credit. GitHub accounts older than 30 days start with $5 of credit. Everyone else, including Google sign-ins, unlocks the same credit by verifying a card, with no charge. Until your first purchase, you can have 20 agents busy at once and spend up to $1 an hour. You can set spend limits per agent and per run.

You can run thousands of agents at once without talking to sales. Before your first credit purchase, an account can have 20 agents busy at once, and the limit rises as your purchases add up.

camelRun works with models from Anthropic, OpenAI, OpenRouter, and Amazon Bedrock. It also works with any server that speaks OpenAI's or Anthropic's API, such as vLLM or Ollama. The default model is Claude Sonnet 5.5.

Not always. You can set up an agent in the console when its tools come from MCP servers, OpenAPI specs, or built-in tools. Tools that run in your own code need the TypeScript or Python SDK.

Every tool call has a deadline. Tools in your own process get 15 seconds without an answer by default. If a call fails or times out, the model is told, and the agent carries on instead of getting stuck.

The model can write and run JavaScript that calls the agent's tools and reads its files. The sandbox is not a Linux machine, so there is no shell and no package installs. Each run gets up to 120 seconds and 32 MiB of memory.

Yes. camelRun is open source under the AGPL. Self-hosting takes one Docker container and Postgres, and it does not need AWS.

Run your first AI agent in five minutes

Connect your coding agent with no key, or sign in with GitHub or Google, create an API key, and run the quickstart.

GitHub accounts older than 30 days start with $5 of free credit. Everyone else unlocks it by verifying a card, with no charge.

Get an API key Read the quickstart
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.

claude mcp add --transport http camelrun https://run.camelai.com/mcp
Docs for coding agents