Skip to content
OpenChamber
↑↓Navigate ↵Open escClose

Classification Providers (Jev)

Some OpenChamber features need a quick decision before they act: is this command risky, what kind of task is this prompt, did the user just start real work? A small decision model called Jev (made by TypeSafe) answers those questions. This page shows how to give OpenChamber access to Jev.

Jev never writes code or replies in your chat. It only returns a verdict, and OpenChamber acts on it. OpenChamber waits at most four seconds for an answer.

What uses Jev

FeatureThe question Jev answers
Permissions & safety netShould the agent ask you before running this action?
Auto model routingWhich of your categories does this prompt belong to?
Sessions in workDid you just ask for real work? Does this turn look like the end of it?
Session recap and suggestionIs a recap or a suggested next message worth generating after this turn?

Without a provider, the safety net and Auto are hidden, and sessions move into work only when you move them. Recaps and suggestions still work, just without the filter.

Set it up

Jev is off until you turn it on: nothing from your conversations goes to a classification provider unless you pick one. Open Settings → Providers → Classification providers and pick where Jev requests go. Besides Off, there are six options.

OpenCode Zen promotion

Free for a limited time, no key needed. OpenCode lets OpenChamber use Jev through OpenCode Zen at no cost while the promotion runs.

When the promotion ends, the option disappears and OpenChamber switches to one of the keys below if you have one.

OpenCode Zen key

Uses the Zen API key saved in OpenCode, paid from your Zen balance.

  1. Open Settings → Providers and add an API key to the OpenCode Zen provider (see Providers, Models & Agents).
  2. Come back to Classification providers and pick OpenCode Zen key.

Signing in with an OpenCode account is not enough. Zen accepts only an API key for Jev.

OpenRouter key

Uses the OpenRouter API key saved in OpenCode, billed to your OpenRouter account. If you already use OpenRouter models in OpenCode, there’s nothing to set up. A key in the OPENROUTER_API_KEY environment variable works too. If both are there, the key saved in OpenCode wins.

  1. Open Settings → Providers and add an API key to the OpenRouter provider.
  2. Come back to Classification providers and pick OpenRouter key.

Vercel AI Gateway key

Uses the AI Gateway API key saved in OpenCode, billed to your Vercel account. A key in the AI_GATEWAY_API_KEY environment variable works too. If both are there, the key saved in OpenCode wins.

  1. Open Settings → Providers and add an API key to the Vercel AI Gateway provider.
  2. Come back to Classification providers and pick Vercel AI Gateway key.

TypeSafe API key

Your own TypeSafe key and quota.

  1. Get an API key from TypeSafe.
  2. Paste it into the TypeSafe API key field and press Save.

Saving a key also selects it. The key stays on the OpenChamber server in its own file and is never sent to the browser.

Custom endpoint

Any service that speaks TypeSafe’s System One API: a self-hosted gateway, a company proxy, or a provider that isn’t listed here.

  1. Under Custom endpoint, fill in the Endpoint URL and the Model (for example jev-latest).
  2. Add an API key if the service needs one. It’s sent as a Bearer token.
  3. Press Save.

Paste the full address the service answers on, like https://api.typesafe.ai/v1/systemone or a company proxy’s https://proxy.company.internal/v1/decision. OpenChamber calls it exactly as you paste it and adds nothing to the path. Only http and https work, and the key goes in its own field, never in the URL.

Saving also selects the endpoint. Like the TypeSafe key, the endpoint and its key stay on the OpenChamber server in their own file, and the key is never sent to the browser.

A server can also get its endpoint from the environment, which is handy when you set up OpenChamber for a team:

Terminal window
OPENCHAMBER_JEV_URL=https://jev.example.com/v1/systemone
OPENCHAMBER_JEV_MODEL=jev-latest        # optional, this is the default
OPENCHAMBER_JEV_API_KEY=...             # optional

On laptops you manage, put the same endpoint under jev in the machine policy file instead. It wins over the environment.

Either way, it replaces the endpoint saved in Settings, and Settings shows it without letting anyone edit it. In enterprise mode it’s the one Jev source allowed besides Off, and it’s selected by default.

The card on Settings → Providers shows Ready once a provider works.

When your pick stops working

If the provider you picked can’t be used (the promotion ended, a key was removed), OpenChamber falls back to the first of your own keys that works: TypeSafe key, then your custom endpoint, then OpenRouter, then Vercel AI Gateway, then Zen key. It never falls back to the promotion, and Off stays off. The page tells you which one it uses right now.

If Jev is slow or fails on a single request, nothing breaks. Each feature has a safe answer for that case:

  • The safety net keeps the action waiting for you.
  • Auto sends the prompt to your fallback model.
  • Sessions in work stay as they are.
  • The recap and suggestion are generated as if Jev didn’t exist.

What Jev sees

Jev gets short excerpts of your conversation: your latest message, and for context up to three previous turns, cut down to their first few hundred characters. For the safety net it also gets the action being checked, the way OpenCode describes it: the command the agent wants to run, or the file and the change for an edit. It never gets whole files or images.

Requests go from the OpenChamber server straight to OpenCode Zen, OpenRouter, Vercel AI Gateway, TypeSafe or your custom endpoint, depending on your pick.

Where it works

Jev runs in the OpenChamber server, so it works in the desktop app, the web app and on your phone connected to either. The VS Code extension talks to OpenCode directly and has no Jev: there’s no safety net, no Auto, and sessions move into work only by hand.