Empryo
config.json. Empryo deliberately keeps credentials out of config files, so instead of writing the token Prism declares the environment variable PRISM_API_KEY and shows you the one command that stores the key.Last updated Reviewed against Prism v0.3.26
Config location
| Platform | Path |
|---|---|
| Windows | %LOCALAPPDATA%\Empryo\config.json |
| macOS | ~/.empryo/config.json |
| Linux | ~/.empryo/config.json |
%LOCALAPPDATA%\Empryo on Windows and does not read the Unix-style path as its global config. An entry written there would only act as a per-project config, which stays inert unless Empryo is started from that exact directory and the directory is trusted — so a provider written there would never reach the model picker.One-click setup
Open the Agents tab
Agents, then Empryo.Click Setup
Store the token
empryo --set-key prism prism
prism.Pick a model
<provider id>/<model id>, so Prism's models appear as prism/<model>.What Prism writes
{
"providers": [
{
"id": "prism",
"name": "Prism",
"baseURL": "http://127.0.0.1:11434/v1",
"envVar": "PRISM_API_KEY",
"reasoning": { "auto": true },
"models": [
{
"id": "glm-5.1:cloud",
"name": "GLM 5.1 (Ollama Cloud)",
"contextWindow": 128000,
"maxOutputTokens": 16384
}
]
}
]
}| Field | Value |
|---|---|
| id | prism. Empryo never replaces a built-in provider whose id collides. |
| baseURL | The Prism proxy's OpenAI-compatible base. |
| envVar | PRISM_API_KEY — the variable Empryo reads the token from and sends as a bearer token. |
| reasoning.auto | true. Reasoning is provider-wide in Empryo, so auto lets Empryo send the thinking fields each model's vendor expects. |
| models[].contextWindow / maxOutputTokens | Your configured values, defaulting to 128000 and 16384. |
Where Empryo keeps the key
Empryo refuses to store credentials in config files. It uses the system keychain on macOS, DPAPI on Windows, and a secret store or ~/.empryo/secrets.json on Linux. Prism therefore never writes a secret into config.json — it only declares which environment variable to read and shows you the setup command.
Why reasoning is set to auto
Empryo has one reasoning setting per provider, not per model. With auto, Empryo looks each model up in models.dev and sends the thinking fields that model's own vendor expects, which is more reliable than Prism guessing field names for every upstream.
Add MCP servers
Empryo stores MCP servers in an array inside the same config file. Prism writes its gateway entry there, and removing access deletes only that entry. See MCP Gateway.
providers list, the key is dropped entirely so the file returns to its original shape.