From a capability
to finished work.

Start in the console

1. Create an account. New accounts receive three run credits.
2. Open Explore and choose a connected capability, or describe a task in Sandbox.
3. Run it, follow Activity and inspect the output.
4. Download the files or save the instructions as an agent.

Your first API request

Create a key under API keys in the workspace. Keep the key on your server, not in browser code. Replace the requestId with a new UUID for each new run. Reusing the same ID and input safely retrieves the same run without a second charge.

curl https://allfurdfr.cloud/api/v1/runs \
  -H "Authorization: Bearer $ALLFURDFR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"requestId":"<new UUID>","capability":"github","input":"vercel/next.js"}'

The API returns HTTP 202 with an id and status. Poll the run until its status is completed, failed or cancelled.

curl https://allfurdfr.cloud/api/v1/runs/RUN_ID \
  -H "Authorization: Bearer $ALLFURDFR_API_KEY"

A completed response includes result, files, events, token usage and timestamps. JSON file contents are returned as strings inside the files object.

Input shapes

{
  "requestId": "<new UUID>",
  "capability": "papers",
  "input": "solar cell efficiency"
}

{
  "requestId": "<new UUID>",
  "prompt": "Write and test a Python unit converter."
}

{
  "requestId": "<new UUID>",
  "agent_id": "<your agent UUID>",
  "input": "Additional context for this run"
}

Errors

401 means missing or revoked credentials. 404 means the resource does not exist or is not yours. 409 indicates insufficient credits, concurrency limits or a conflicting request ID. 429 is the API key request limit. Errors return a JSON error message. Do not retry a new ID unless you intend a new run.

Reusable agents and schedules

Save a name, instructions and an optional public repository workspace. Run the agent manually from the console, call it through the API or use the MCP submit_run tool. Pause an agent to prevent new runs. Active subscribers can enable daily or weekly schedules; the hourly dispatcher submits due work subject to credits and capacity.

Templates are starting instructions, not guarantees. Add a concrete topic or task before running. Saved agents are private to your account and callable only with your credentials.

A fresh environment for every task

Sandbox tasks run in a hosted Linux environment with Node and Python. The agent can read connected public data, write files and execute commands. A workspace can supply a public GitHub repository. Each run starts fresh; later runs do not automatically inherit earlier edits.

After the run, supported text files are saved and the sandbox stops. Download the archive and review it before use. Large binary files, environment files, Git internals and dependency directories are excluded.

Limits and credits

Paid plan credits are granted only after a verified paid invoice. Purchases are not available yet. Starter runs are available without a card.