JernCloudStart with GitHubJern Cloud docs
Provider keys.
Sessions run on your workspace's own model provider key. Jern holds the key encrypted, forwards each request on the agent's behalf, and meters usage from the provider's own report.
Supported providers
| Provider | Models offered in the session dialog | Notes |
|---|---|---|
| Anthropic | Claude Opus 5, Claude Sonnet 4.5 | Relayed natively. |
| OpenAI | GPT-5.5, GPT-5.4 | Reasoning tokens count toward the cap, so default caps start higher. The Codex models answer only on OpenAI's Responses API and are not offered yet. |
| Gemini | Gemini 3.8 Flash, Gemini 3.1 Pro Preview | Reached through Google's OpenAI-compatible endpoint. Thinking counts toward the cap. |
| Fireworks | DeepSeek V4 Pro, Kimi K2.7 Code, GLM 5.3, Qwen 3.8 Max, GPT OSS 120B | Open-weight models on Fireworks serverless at a fraction of frontier prices. They reason in the open, and reasoning counts toward the cap. |
A workspace holds one active key per provider. The session dialog groups models by provider and says which providers have a key.
Adding, rotating, and revoking
Under Settings › Provider, paste a key. It is validated once with the provider's models endpoint, encrypted before storage with keys held in AWS KMS, and never shown again. Rotating stores a new version; attempts already pinned to the old version fail closed rather than falling back. Revoking a key stops new attempts on that provider until another is added.
What the agent sees
Nothing. The agent runs in a private network namespace whose only route is a relay to the Jern gateway, and it authenticates to the gateway with a token scoped to one run. The gateway adds your provider key and forwards the request; what the provider returns is what the agent gets.
Metering and caps
Every attempt has a token cap chosen when the session opens. The gateway lowers the request's own output limit to the remaining budget, reads input and output tokens from the provider's usage report, and refuses further requests once the cap is spent. Counts are in the provider's own tokens and are not comparable across providers. Usage per attempt and per month appears under Settings › Usage.
Billing
The provider bills you directly for tokens. Jern's price covers the control plane only.