AI agents
Self-hosted AI agent sandboxes
An agent that can actually see what broke.
Temps runs AI-agent sandboxes on your own server, as Docker containers or Firecracker microVMs: a self-hosted alternative to E2B, free to self-host with no per-second billing. Next to them sit an AI chat that reads your own traces and metrics and an OpenAI-compatible AI gateway, all wired into the telemetry every other feature already collects.
Bring your own provider keys
Ask questions about a checkout latency spike and get answers pulled from your own traces, metrics, and revenue data — not a generic assistant guessing from a prompt.
Key features
It reads your telemetry, not a generic prompt
The AI chat pulls from the same traces, metrics, and error data every other Temps feature already collects — ask about a specific spike and get an answer grounded in what actually happened.
A real sandbox, not a container ID
Agent sandboxes run on a Docker/Firecracker backend with an in-browser command runner and password-protected preview URLs — a workspace you can actually open and watch, not a black box.
One gateway, any provider
The AI gateway sits behind a single OpenAI-compatible endpoint. Bring your own provider keys and swap models without rewriting integration code.
Usage tracked per project
Token and request usage is visible per project from day one — no separate dashboard to check.
Scoped deployment tokens for CI and agents
Agents and CI pipelines authenticate with role-scoped API keys, so automated deploys get exactly the access they need and nothing more.
Hand an error straight to an agent
When something breaks, the same error tracking and tracing data that surfaces the problem can hand it directly to a sandboxed agent already reading your codebase.
Bring your own provider keys
Temps vs E2B
Both give an agent an isolated Linux environment to run code in. E2B runs every sandbox as a Firecracker microVM in its cloud. Temps runs sandboxes on a server you own: as Docker containers by default, with capabilities dropped and no-new-privileges set, or as Firecracker microVMs with their own guest kernel once a Linux host with KVM has been provisioned for it.
| Dimension | Temps | E2B |
|---|---|---|
| Where sandboxes run | Your own server | E2B's cloud; BYOC (Enterprise) or E2B Embed to run it yourself |
| Isolation | Docker container by default; Firecracker microVM, chosen per sandbox, on hosts with /dev/kvm | Firecracker microVM for every sandbox |
| Billing | Free to self-host; you pay only for your server | Per second: $0.000014 per vCPU and $0.0000045 per GiB of RAM; Pro $150/month plus usage |
| Long-running work | Workspaces suspend when idle and wake on the next command | Hobby sessions last up to 1 hour |
E2B is the more mature product, with published SDKs and agent-framework integrations, while Temps' Node sandbox SDK is in the Temps repository but not yet on npm. The E2B alternatives guide compares the wider field.
Competitor prices and plans verified on ; check each vendor's pricing page before you buy.
Questions
Further reading
Is Temps a self-hosted alternative to E2B?
Yes, if you can run your own Linux server. Temps sandboxes are isolated environments you drive from the CLI or the REST API, running as Docker containers by default or as Firecracker microVMs on hosts with KVM. E2B is a managed service billed per second; Temps is free to self-host, so sandboxes cost only the server they run on. E2B's SDKs and integrations are more mature.
How are Temps sandboxes isolated?
By default each sandbox is a Docker container with Linux capabilities dropped and no-new-privileges set, sharing the host kernel. On a Linux host with /dev/kvm, temps firecracker setup enables a Firecracker backend that boots each sandbox as a microVM with its own guest kernel. You pick the backend per sandbox, and asking for one the host can't provide fails instead of quietly falling back to a container.
How much do agent sandboxes cost on Temps?
Nothing beyond your server. Temps is free to self-host and has no per-second, per-sandbox or per-event pricing; sandboxes use the CPU, memory and disk of the machine they run on. Model calls through the AI gateway are billed by the provider whose API keys you bring.
What does the host need to run sandboxes?
Docker, which Temps already requires. The Firecracker backend additionally needs Linux on x86_64 or aarch64 with hardware virtualisation (/dev/kvm) and a one-time temps firecracker setup. macOS hosts can't enable the Firecracker backend.
Give your agents something real to work with.
Install Temps, deploy a project, and the sandbox, gateway, and telemetry are already there — no separate service to wire up.
Try it yourself — quickstart guide