Solutions

Stackap for production AI apps

Stackap hosts the application around a model, its web app, API keys, database and scheduled jobs, but it does not host models, GPUs or background workers.

The situation

An AI app, in the sense most people mean, is an ordinary web app that calls a model provider's API. The model runs elsewhere. What you need from hosting is what any app needs, plus extra care for the API keys, which are the most expensive secrets you hold.

That is the part Stackap covers. If you mean running models yourself, on your own GPUs, this is the wrong platform.

What Stackap gives you here

Encrypted, masked storage for provider API keys, set from standard input so they never reach a transcript or a shell history.
Build-time protection: non-public values are placeholders during the build, so a key is never baked into an image.
Runtime and build logs your agent or you can read to see why a request failed.
Rollback, for the day a prompt or model change makes the app worse.
Cron jobs for scheduled work such as evaluations, digests or clean-up.
A Postgres database for conversations, results and usage records.

How it looks in practice

Keep each provider key in its own variable, rotate by setting a new value, and give development and production different keys. Log what you need to debug a failure but not the raw prompts of your users.

Treat cost as an application concern. Stackap shows request counts and load time, not spend per provider, so record token usage in your own database.

Early-access fit

Early access is a less natural fit if you need GPUs, or to run open models yourself. Stackap runs ordinary containers on one CPU server.

Early access is a less natural fit if your work arrives as jobs that run for minutes with no web request. A worker needs an HTTP port to pass the health check, and there is no queue.

Early-access scope

No GPUs and no model hosting.
No queue and no background workers without an HTTP port.
Very long or streaming responses have not been tested end to end through the proxy, so test yours.
No usage or spend tracking for providers.

Questions

Where should my keys live?

In environment variables set from standard input. Public prefixed variables are readable by every visitor, so never put a provider key in one.

Can a request run for a long time?

Stackap runs your app as an ordinary container, not as short serverless functions, but we have not measured limits on long or streaming responses, so test the longest one you will serve.

Stackap is in early access. Tell us what you run and we will reply with a straight answer about whether it fits.

Ask for an invitation

Last updated 2026-10-03.