Limits
These are the limits in force today. Most belong to a vendor or to the one node that runs rent.dev's own sandboxes, and they are passed through as they are. They describe the current setup and will move when it does.
https://api.rent.dev/The current node
rent.dev's own sandboxes run on one node, which takes up to 200 running sandboxes at once. That number is a cap, not the node's capacity: how many more it can start follows the memory still free on it. The node's health and its free slots are in the status answer above and on the status page. Free slots are the cap less the sandboxes running, or the number of the smallest template its free memory still fits, whichever is smaller.
| Limit | Value |
|---|---|
| Cap on sandboxes running at once, across all accounts | 200 |
| Longest session | 24 h |
| Free disk per sandbox, beyond the template's image | 512 MiB |
When the cap is reached, or the node has less free memory than the template needs, a create answers 503 no_capacity.
Templates
| Template | vCPU | Memory |
|---|---|---|
base | 2 | 512 MiB |
code-interpreter-v1 | 2 | 2 GiB |
- A sandbox takes its template's size.
cpuCounton an E2B-style create must equal it. - You can build your own templates from an image or from another template, with command steps.
- A build cannot upload files, so a
copy()step is refused. Fetch what the build needs from inside a command instead.
Requests
| Limit | Value |
|---|---|
| Request body, including a file write | 100 MB. The edge refuses a larger request body. |
| Sandbox timeout | Default 300 s, at most 24 h from now. |
exec and run-code timeout | Default 50 s, at most 3600 s. Start a process for longer work. |
Prefer: wait on create | Up to 999 s. |
wait_ms when reading process output | Up to 30000 ms. |
| List page size | 1 to 200. |
| Quote shape | 1 to 64 vCPU, memory from 128 MiB. |
| Idempotency replay | 24 hours. |
| Pay link | 24 hours. |
GPU pods
| Limit | Value |
|---|---|
exec timeout | 1 to 25 s. RunPod's limit. |
exec output | 1 MiB. RunPod's limit. |
Stock is RunPod's. A create for a GPU type that is out of stock fails.
A GPU pod has no timeout, and no call pauses or resumes one. It runs, and RunPod bills it, until you terminate it.
Not available
- Volumes. A sandbox's disk lasts as long as the sandbox, or its snapshot.
- Sandbox metrics (
getMetrics). - Template file uploads (
copy()steps). - Workload identity (
iam), the E2B MCP gateway (mcp) and an egress proxy. - A filesystem-only pause. A pause always keeps memory.
- Pause, resume, snapshot, fork and a log stream for a sandbox on your own Vercel token; a snapshot of a sandbox on your own E2B key; pause and resume for a GPU pod. What each vendor can do has the table.
The full list, call by call, is the E2B compatibility matrix.
Rate limits
A 429 rate_limited is a vendor's own rate limit, passed through with its Retry-After. Wait that long and send the request again.