IN THE FIELD GUIDE
coreloop
coreloop.run
Coreloop is an open implementation of the Claude Managed Agents API that runs agents in sandboxed containers with durable sessions and message queuing. It supports any model billed to the user's own provider keys and allows streaming of output events.
THE PRODUCT, BEYOND THE PITCH
Editorially reviewed · Sources checked Sep 10, 2026
A good fit for
- Developers needing a managed agent platform compatible with Claude Managed Agents API and multiple AI providers.
- Users who want to run agents with their own API keys for billing across OpenAI, Anthropic, or OpenRouter models.
Know the limitations
- A session fails at creation with invalid_request_error if no configured key serves its model.
What you can do
- Running AI agents for tasks such as code review with persistent, steerable sessions.
- Migrating from Claude Managed Agents API with minimal code changes.
Features
- Sandboxed execution with isolated containers including bash and filesystem.
- Persistent sessions with stored event history allowing disconnect and resume.
- Event streaming via server-sent events for messages, tool calls, and status changes.
- Support for MCP servers to connect user tools.
- Ability to steer agents mid-run by sending messages that are read on the next turn.
Integrations
- Compatible with Anthropic SDKs by changing the base URL to coreloop's endpoint.
Platforms & data export
- Runs agents in managed infrastructure with sandboxed containers supporting bash and filesystem.
THE COST FOR YOUR TEAM
Go beyond the starting price.
Published plan prices for your team size and usage. Results update as you type. Taxes, currency conversion and unlisted add-ons are excluded, and anything the source did not state is called out rather than guessed.
Known monthly subtotal
$0.00/month
1 of 1 tools could not be priced with these inputs, so this is not the full cost.
| Tool / plan | Monthly | Per year | What this assumes |
|---|---|---|---|
| No pricing recorded yet. Check the official site, or ask the owner to add it. | |||
A practical workflow
- Create an agent with a name, model, system prompt, and tools.
- Start a session with the agent and environment, sending initial user messages.
- Read the event stream to receive messages, tool calls, and status updates in real time.
Based on the sources below. Editorial review does not imply hands-on product testing.
Alternatives to explore
Filter alternatives →Candidates based on category and primary feature. Check feature and pricing differences before switching.
Plan a switch from coreloop →What changed
Changes to the facts recorded here, not a live scan of every vendor update. Save this tool to follow updates in your account.
No changes recorded yet.