Introduction
Isolated sandboxes for untrusted code — persistent sessions, full Linux, CLI, SDKs, and MCP.
inis.run is a code-execution sandbox. You call an API (or the CLI), each run gets its own isolated Linux machine — shell, files, packages, preview URLs — and persistent sessions when you need state across tool calls.
How to connect
Same API behind every surface. Pick what fits:
| Surface | Install | Best for |
|---|---|---|
| CLI | curl -fsSL https://inis.run/install.sh | bash | Scripts, CI, quick manual runs |
| HTTP API | Bearer token | Any language, custom integrations |
| Python SDK | pip install inis | Python agents, data pipelines |
| TypeScript SDK | npm install @inis-run/sdk | Next.js, Node agents |
| Remote MCP | https://mcp.inis.run/sse + API key | Claude, Cursor, cloud-hosted agents |
Framework helpers: Pydantic AI, AI SDK, LangChain, OpenAI Agents SDK, Strands Agents, Mastra, Google ADK, Microsoft Agent Framework.
What you get
- Persistent sessions — state survives across tool calls; auto-pause when idle, wake on the next op
- Full Linux — arbitrary shell, file I/O, HTTPS preview URLs
- Templates, fork, checkpoints — start from a known env, branch for parallel work, roll back when you need to
Next steps
- Quickstart — install, sign in, run your first command
- Concepts — sessions, previews, fork, checkpoints
- Session lifecycle — templates, live/paused, hot/warm/cold resume
- Production best practices — lifecycle knobs, bulk upload, idempotent creates, reconnecting after resume
- Building agents — Claude, Cursor, or any remote MCP client
- llms.txt — curated docs index for agents