This is an example of how an application to use an LLM can be configured to easily customize the system instructions and enabled tools using Optify.
See Configurable Lists and Configurable Strings for more details on how to dynamically build lists and strings.
Other properties such as the tool input schema can also be customized adjacent to the tool's description.
Here is a simple configuration. Imagine it was the result of merging multiple configuration files based on several features.
Here is how your code would see the configuration when using an OptionsProvider with configurable values enabled:
{
"options": {
"model": {
"name": "GPT-5",
"parameters": {
"max_tokens": 1500,
"temperature": 0.7
}
},
"chat": {
"system_instructions": "You are a helpful assistant for MyApp.\n\nYou are nice.",
"tools": [
{
"name": "code_executor",
"description": "Executes code snippets in a secure environment."
},
{
"name": "web_search",
"description": "Search the web using Bing Search."
}
]
}
}
}A custom schema at .optify/schema.json can specify the value type for tools while continuing to use Optify's standard definitions:
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"additionalProperties": false,
"minProperties": 1,
"properties": {
"conditions": {
"$ref": "https://raw.githubusercontent.com/juharris/optify/refs/heads/main/schemas/feature_file.json#/definitions/conditions"
},
"imports": {
"$ref": "https://raw.githubusercontent.com/juharris/optify/refs/heads/main/schemas/feature_file.json#/definitions/imports"
},
"metadata": {
"$ref": "https://raw.githubusercontent.com/juharris/optify/refs/heads/main/schemas/feature_file.json#/definitions/metadata"
},
"options": {
"type": "object",
"additionalProperties": false,
"minProperties": 1,
"properties": {
"model": {
"type": "object",
"additionalProperties": false,
"properties": {
"name": {
"type": "string"
},
"parameters": {
"type": "object"
}
}
},
"chat": {
"type": "object",
"additionalProperties": false,
"properties": {
"system_instructions": {
"$ref": "https://raw.githubusercontent.com/juharris/optify/refs/heads/main/schemas/feature_file.json#/definitions/configurableString"
},
"tools": {
"allOf": [
{
"$ref": "https://raw.githubusercontent.com/juharris/optify/refs/heads/main/schemas/feature_file.json#/definitions/configurableList"
}
],
"items": {
"type": "object",
"additionalProperties": false,
"properties": {
"description": {
"$ref": "https://raw.githubusercontent.com/juharris/optify/refs/heads/main/schemas/feature_file.json#/definitions/configurableString"
},
"name": {
"type": "string"
}
}
},
"additionalProperties": {
"properties": {
"$value": {
"$ref": "#/properties/options/properties/tools/items"
}
}
}
}
}
}
}
}
}
}This validates both literal tool arrays and configurable tool entries while providing editor suggestions for "name", "description", and the nested ConfigurableString properties.
The tool schema is defined once under "items" and reused for each configurable entry's "$value".
"name" and "description" intentionally remain optional because each feature file is validated before features are merged, and a feature may only set one property.
{ "options": { "model": { "name": "GPT-5", "parameters": { "max_tokens": 1500, "temperature": 0.7 } }, "chat": { "system_instructions": { "$type": "Optify.ConfigurableString", "base": { "file": "templates/system_instructions.liquid" }, "arguments": { "product_name": "MyApp", "personality": { "file": "shared/personality.txt" } } }, "tools": { "$type": "Optify.ConfigurableList", "code_executor": { "$value": { "name": "code_executor", "description": { "$type": "Optify.ConfigurableString", "base": "Executes code snippets in a secure environment." } } }, // Map to `null` to disable a tool. "math": null, "web_search": { "$value": { "name": "web_search", "description": { "$type": "Optify.ConfigurableString", "base": { "liquid": "Search the web using {{ provider }}." }, "arguments": { "provider": "Bing Search" } } } } } } } }