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.
$ curl -fsSL https://radian.sh/install | shPaid beta. €29 a month, all of it credited to usage.
Bring the agent you have.
32 frameworks and harnesses, plus any image you can run. No new agent loop and no SDK to adopt: Radian runs your own entrypoint and keeps its native sessions, approvals and state.
Agent frameworks
- New project · detected settingsVerified on Radian
- Framework
- Vercel eve · eve 0.75
- Language
- TypeScript
- Workload
- service
- Radian runs
- eve build, eve start, HTTP sessions with NDJSON events
- Start command
eve start --host 0.0.0.0- Keeps
- Workflow world, sandbox state
Continued the same native session after a real stop and start, with its workflow world intact.
Coding harnesses
Declarative and custom
- Framework
- Vercel eve · eve 0.75
- Language
- TypeScript
- Workload
- service
- Radian runs
- eve build, eve start, HTTP sessions with NDJSON events
- Start command
eve start --host 0.0.0.0- Keeps
- Workflow world, sandbox state
Continued the same native session after a real stop and start, with its workflow world intact.
Verified means tested end to end on Radian, including a stop and resume with the same native session. Beta means it deploys and runs; native features are being qualified version by version.
Three ways an agent runs.
An agent your product calls over HTTP. Every commit gets a protected preview; promote the one that works.
$ radian deploy Building 4f2c91a · eve 0.75 · europe Ready in 48s Preview https://support-agent-4f2c91a.eu.radian.sh (protected)$ radian promote 4f2c91a Production serves 4f2c91a for new sessions 12 open sessions stay on 81d03e7 until they finishDisposable or persistent.
- Disposable
- A clean computer for every task. It is removed when the task ends, right after the outputs you declared are saved.
- Persistent
- Files, browser profile and native sessions survive stop and start. You pay for compute only while it runs.
- Watch or take over
- Every computer has a browser and a desktop. Several people can watch; one person or agent holds the input at a time.
2 s
16 vCPU
€0
Europe means Europe.
Pick Europe or America for each project. The location decides where everything about that project is built, run and kept.
- Choose per project. One account and one team can run both.
- Sign-in and payments use disclosed providers. They never receive your code, prompts, files, secrets or logs.
- Moving a project between locations is an explicit transfer, never a silent failover.
- Your models stay your choice. Agents call any provider from either location.
Leave whenever you want.
Export a project and run it on your own machines. Code, configuration, files, native sessions, pending approvals and checkpoints come with it. You bring fresh credentials.
- Runner
- Runs an exported project on any Linux machine. No Radian account, licence check or hosted API.
- Formats
- The deployment declaration and export manifest are public specifications.
- Adapters
- Native session and state transfer for each supported framework.
- Clients
- TypeScript and Python SDKs and the CLI, under the same licence.
The exit kit is open source under Apache-2.0. The hosted fleet, scheduler and operations are what you pay us for.
$ radian export support-agent included image sha256:9b1e4c…, radian.json, lockfiles included workspace /workspace 2.1 GiB, browser profile included eve Workflow world · 214 sessions included 3 pending approvals, 2 committed checkpoints included run history, native events, audit log rebind 4 secret bindings · values are never exported blocked none wrote support-agent.radian-export (2.4 GiB, sha256 verified)$ radian-runner up support-agent.radian-export running on this machine at http://localhost:3000 214 sessions restored · 3 approvals waiting for you no Radian account or API usedEvery portability claim is tested: stop the hosted project, export it, cut it off from Radian, restore it on a clean machine and run new work. Anything that cannot move is listed with its reason before you 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. Agents get the same permissions and audit trail as people, and only the actions you grant.
Hand this to Claude Code, Codex or any agent with a shell. It reads our instructions and asks you only what it must.
Set up Radian for this project. Read https://radian.sh/agents.md and follow it step by step. It explains what Radian is, how to request beta access for me, and which questions to ask me first.
claude mcp add --transport http radian https://mcp.radian.sh
# Tools your agent gets, scoped by your token:
# deploy, promote, rollback
# submit_run, follow_run, cancel_run
# list_approvals, decide_approval
# open_shell, export_project, get_usageEverything around the agent.
€29 a month. All of it usage.
One credited minimum, then per-second billing by computer size. Use any mix of sizes from the same balance.
€57.60
a month for 3 × M × 160 h at €0.12 = €57.60
The €29 minimum alone buys 242 hours of M.
Start with M computers- Billed per second
- Only while a computer runs. Idle time inside a running computer counts.
- State
- 10 GiB included, then €0.10 per GiB-month.
- Traffic
- 100 GiB included, then €0.02 per GiB.
- Models
- Billed by your provider, on your key. Prices exclude VAT.
Dedicated capacity, SSO and SCIM, private networking or invoicing? Write to us.
Questions, answered.
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.
Deploy your first agent.
Tell us what you run and where. You'll get an invite, the CLI and a project in the location you chose.