An open-source agent platform from LangGenius, the team behind Dify. It runs coding agents such as the Claude Agent SDK (Claude Code), OpenAI Codex, and OpenCode inside isolated sandboxes and exposes them through a unified HTTP API. Its role is to lift an agent you have been running locally into an API your application can call, with three entry points: a web console, a public API, and a CLI. You can self-host it on your own Cloudflare account and keep your model API keys yourself (BYOK). It is licensed under Apache 2.0, and v0.1.0 was released as an alpha in July 2026.
Key Features
- Multiple coding agents behind one API: A common runtime protocol absorbs the differences between harnesses such as the Claude Agent SDK, OpenAI Codex, and OpenCode, so the calling side works with the same API
- Execution in isolated sandboxes: Agents run inside sandboxes, and tool-execution event streaming and permission prompts can also be handled through the API
- Durable Threads and Runs: Threads for conversations, Runs for executions, and generated files persist across sessions, so you can reconnect to interrupted work and continue it
- Self-hosting on your own Cloudflare account: The control plane runs on Cloudflare resources such as Workers, D1, and R2, with dedicated migration and deployment paths for production
- App-level operational view: Providers, models, skills, MCP servers, files, run history, usage costs, and deployments are all managed within an “App” boundary
- Skills and MCP server support: You can package procedures as Skills for an agent, or connect to external tools through MCP servers
Pricing
| Plan | Price | What it covers |
|---|---|---|
| Self-hosted (OSS) | The software itself is free (Apache 2.0) | Deployed to your own Cloudflare account. Actual costs are Cloudflare usage fees and model provider API fees (BYOK) |
| Hosted (cloud.mosoo.ai) | Check the official site | Offered as the public API endpoint. No published pricing information could be confirmed at the time of writing |
Pricing is current as of August 2026. Please check the official site for the latest pricing.
Pros & Cons
✅ Pros
- Turns a coding agent you tested locally into an API your application can call
- The calling code stays the same across harnesses, making it easier to switch between or combine agents
- Because it is self-hosted, the code, the data, and the model API keys all stay under your control
- Usage and cost can be tracked per App, per agent, and per model
- Apache 2.0 open source, so you can read the implementation to confirm its behavior
⚠️ Cons
- It is at an alpha stage, and the project states that APIs, data models, and deployment methods may change without backward compatibility
- Self-hosting requires a Cloudflare account and operational knowledge; it is not a no-code product
- Local development requires prerequisites such as Bun, Just, and Docker
- Model usage fees and Cloudflare’s usage-based charges are your own responsibility
- Japanese-language documentation and community resources are still limited
Comparison with Similar Services
| Item | Mosoo | Dify | E2B | Cloudflare Agents SDK |
|---|---|---|---|---|
| Main role | Runtime for coding agents and turning them into APIs | Platform for building LLM apps and workflows | Sandboxes for code execution | SDK for implementing agents on Cloudflare |
| Provider | LangGenius | LangGenius | E2B | Cloudflare |
| Audience | Developers embedding agents into applications | App teams including non-engineers | Developers who only need an execution environment | Developers building on Cloudflare themselves |
| Self-hosting | Yes (Cloudflare) | Yes | Partially | Build-it-yourself by design |
| License | Apache 2.0 | Open-source edition available | Open-source edition available | Open source |
Who Is It For
- Developers who want automation with Claude Code or Codex to be callable from their own app or service
- Teams that want the agent execution environment inside their own cloud account rather than a SaaS
- People who want to compare or combine several coding agents without splitting the calling-side implementation
- Operators who need visibility into usage cost per agent and per model
- Anyone who wants to study the implementation of an open-source agent platform
Summary
Mosoo fills the gap between a coding agent running on your laptop and an API you can embed in a real product. What stands out is that the parts you need once the agent goes into operation ─ sandboxed execution, durable Threads and Runs, and visible usage cost ─ are built in from the start. On the other hand, it is at an alpha stage, so you should expect specification changes. A reasonable path is to run it locally first to see the shape of the API, then consider deploying it to your own Cloudflare account.