# Radian: Deploy the agent you already built.

Radian hosts your existing agent framework or coding harness in isolated computers, in Europe or America. You keep your code, your models, and a tested way to run it elsewhere.

Paid beta. €29 a month, all of it credited to usage. Start: https://radian.sh/start

## Bring the agent you have

No new agent loop and no SDK to adopt. Radian builds your repository or image, runs its own entrypoint and keeps its native sessions, approvals and state.

| Framework | Language | Radian runs | Keeps | Status |
| --- | --- | --- | --- | --- |
| LangChain | Python | create_agent() graph, invoke and stream | checkpointer threads, tool events | Verified |
| LangGraph | Python, TypeScript | compiled StateGraph, Command(resume=…) | checkpoints, cross-thread store | Beta |
| Deep Agents | Python | create_deep_agent() on the LangGraph runtime | graph state and sandbox backend files | Beta |
| CrewAI | Python | Crew.kickoff() and Flows | flow state, memory and knowledge stores | Beta |
| Pydantic AI | Python | Agent.run(), run_stream_events() | message history, deferred approval calls | Beta |
| OpenAI Agents SDK | Python, TypeScript | Runner.run_streamed(), SandboxAgent | RunState, sessions, sandbox snapshots | Beta |
| Claude Agent SDK | Python, TypeScript | query() with permission callbacks | transcripts, session store | Beta |
| Google ADK | Python | Runner.run_async(), adk api_server | session, artifact and memory services | Beta |
| Microsoft Agent Framework | Python, .NET | agents and graph workflows with checkpoints | threads, workflow checkpoints | Beta |
| LlamaIndex | Python, TypeScript | AgentWorkflow and FunctionAgent with streamed events | workflow context, chat memory | Beta |
| Agno | Python | Agent.run() and AgentOS | sessions and memories in its database | Beta |
| Strands Agents | Python, TypeScript | Agent() calls with streamed events | session manager state | Beta |
| smolagents | Python | CodeAgent and ToolCallingAgent | agent memory steps | Beta |
| AG2 | Python | ConversableAgent group chats | chat history | Beta |
| Mastra | TypeScript | Agent.stream(), durable agents with runId | memory, workflow snapshots | Beta |
| Vercel AI SDK | TypeScript | Agent and streamText with tools | messages in your own store | Beta |
| Vercel eve | TypeScript | eve build, eve start, HTTP sessions with NDJSON events | Workflow world, sandbox state | Verified |
| Pi SDK | TypeScript | createAgentSession(), prompt(), subscribe() | session tree via SessionManager | Beta |
| Pi Durable | TypeScript | Harness.open(), resume() | SQLite and JSONL journal, task registry | Beta |
| Claude Code | CLI | headless stream-json, or its own terminal | transcripts and workspace | Beta |
| Codex | CLI | codex exec --json, exec resume | rollout files and config | Beta |
| Droid | CLI | droid exec headless, or its own terminal | sessions and workspace | Beta |
| fx | CLI | fx ask one-shot runs, or fx acp for editors | workspace and config | Beta |
| Pi Harness | CLI | pi --mode rpc over JSONL | branchable session files | Beta |
| Gemini CLI | CLI | headless prompts with streamed JSON, or its own terminal | checkpoints and workspace | Beta |
| OpenCode | CLI | opencode run, or opencode serve as a server | sessions | Beta |
| Goose | CLI | goose run with recipes, or its own terminal | sessions and recipes | Beta |
| Amp | CLI | execute mode, or its own terminal | threads | Beta |
| Cursor CLI | CLI | headless print mode, or its own terminal | chats and workspace | Beta |
| Aider | CLI | aider --message, or its own terminal | chat history in the repository | Beta |
| OpenHands | CLI | headless runs, or its own interface | conversation state | Beta |
| Docker Agent | YAML | docker agent run with root and sub-agents | sessions; MCP tool servers run beside it | Beta |
| Anything else | Any language | your image and start command | the paths you declare | Beta |

Verified means tested end to end on Radian, including a stop and resume with the same native session.

## Three ways an agent runs

- **Services.** An agent your product calls over HTTP. Every commit gets a protected preview; promote the one that works.
- **Background jobs.** Submit work, close the laptop, come back to the same run. Approvals wait for you, not for a timeout.
- **Coding sessions.** Claude Code, Codex or Pi in a hosted workspace. Prompt it, watch it, open its own terminal, or a shell beside it.

## Disposable or persistent

- Disposable: a clean computer for every task, removed when the task ends after declared outputs are saved.
- Persistent: files, browser profile and native sessions survive stop and start; compute is billed only while running.
- Every computer has a browser and a desktop; one person or agent holds the input at a time.
- Attaching to a warm computer with browser and desktop ready measured 1.65 to 2.09 seconds.

## Europe means Europe

Choose Europe or America per project, under one account. In Europe, builds, registries, computers, workspaces, native state, secrets, logs and backups run on European-owned, European-operated infrastructure. Sign-in and payments use disclosed providers that never receive code, prompts, files, secrets or logs. Customers choose their own model providers.

## Leave whenever you want

Export a project and run it on your own machines with the open-source exit kit (Apache-2.0): runner, formats, adapters, SDKs and CLI. No Radian account is needed to run an export.

## Your agents can run Radian too

Everything in the dashboard is an API call, with TypeScript and Python SDKs, a CLI and an MCP server, under the same permissions and audit trail.

## Pricing

| Size | vCPU | Memory | Per hour |
| --- | --- | --- | --- |
| S | 2 | 4 GiB | €0.06 |
| M | 4 | 8 GiB | €0.12 |
| L | 8 | 16 GiB | €0.24 |
| XL | 16 | 32 GiB | €0.48 |

€29 monthly minimum, credited to usage. Per-second billing while running. 10 GiB state included, then €0.10/GiB-month. 100 GiB traffic included, then €0.02/GiB. Prices exclude VAT.

## Questions

### Do I have to rewrite my agent?

No. Radian builds your repository or image and runs its own start command. A native adapter adds sessions, approvals and resume where your framework supports them. Without one you still get builds, previews, isolation, logs and exports.

### Which models can I use?

Any. Your agent calls the model providers you choose, with your keys, from either location. Radian does not proxy, resell or bill inference.

### What does European-owned cover?

The whole workload pipeline of a European project: builds, registries, computers, workspaces, native state, secrets, logs and backups. Sign-in and payments use separately disclosed providers that never receive your code, prompts, files, secrets or logs.

### What happens to running sessions when I deploy?

Promotion changes where new work goes. Open sessions and queued jobs stay on the version they started with until they finish or you move them. Rolling back code does not undo what an agent already did; the run history shows which version did what.

### Can I run my agent without Radian?

Yes. The exit kit is open source under Apache-2.0. Export a project and run it on your own machines with no Radian account: code, configuration, files, native sessions, pending approvals and checkpoints. You bring new credentials. The hosted fleet and scheduler stay ours.

### What is in the paid beta?

Services, background jobs and coding sessions; the dashboard, API, TypeScript and Python SDKs, CLI and MCP server; previews, promotion, rollback, teams, spend limits and exports; Europe and America. SSO, SCIM, private networking and dedicated capacity are switched on per team on request.

### How is it billed?

€29 a month, all credited to usage. Computers bill per second while they run, by size. A stopped computer costs only the state it keeps beyond 10 GiB. Prices exclude VAT.