Reference
Statuses, limits, commands and where everything lives.
Statuses
| Run | Means |
|---|---|
queued |
Created, waiting for a slot. |
running |
An agent is working on it. |
judging |
The agent finished; the evaluator is reading the record and checking the machine. |
passed |
Judged; every criterion met. |
failed |
Judged; at least one criterion not met. |
errored |
Did not complete — crash, missing credential, timeout, failed setup step. |
cancelled |
Stopped by hand. |
| Experiment | Means |
|---|---|
queued |
Launched; nothing has started yet. |
running |
At least one run is outstanding. |
complete |
Nothing of its is queued or running. |
Limits and defaults
| Trials per run | 1–10 (default 1) |
| Max concurrency per experiment | 1–32 (default 4) |
| Time limit for one attempt | 20 minutes by default |
| Time limit for one setup step | 15 minutes |
| Largest single uploaded file | 100 MB by default |
| Evaluator investigation steps | 40, then it must decide on what it established |
| Default machine | Standard — 2 vCPU, 4 GB, 20 GB |
| Repository clone | --depth 1, public repositories only |
The time and size limits are set by whoever runs your Redline; the numbers above are the defaults.
Commands
| Command | What it does |
|---|---|
npx redline init |
Scaffolds the doorway file (and redline.config.ts). |
npx redline dev |
Connects your agents and executes their runs. |
npx redline doctor |
Checks key, URL, agent file, env files and tool injection. |
Python drops the npx. Environment: REDLINE_API_KEY (your runner key,
rl_…), REDLINE_URL (where the platform lives).
Where things live
| Page | What it is for |
|---|---|
| Home | What is running, what is missing, what happened last. |
| Ask Redline | The project agent. |
| New experiment | Write a task and launch. |
| All experiments / All runs | Everything that has happened. |
| Tasks | Instructions, and the files that belong to them. |
| Rubrics | The standards runs are judged against. |
| Files | Repositories and uploads placed on the machine. |
| Machines | Machine shapes and images. |
| Agents | Connect catalog agents; connect your own. |
| Assets | MCP servers, skills, CLIs, SDKs, plugins. |
| Schedules | Experiments that repeat themselves. |
What is stored, and what is not
Provider keys
Encrypted at rest. What leaves the database for the interface is the provider, a hint and when it was connected — never the value. The plaintext travels in one direction only: into a container that is about to start a run. Redline cannot show you a key again; disconnect and reconnect to replace one.
Your own agent's code and keys
Never sent. redline dev connects outbound and executes your function in
your process; what arrives here is the record of what it did.
Run records
Transcripts, final answers, verdicts and notes are kept with the run and deleted with its experiment.
Machines
Created per run and destroyed after judging. Nothing on a machine survives the run that created it.
Keyboard
/ focuses the search box on any page that has one.