How to Configure Multiple LLM Models Under a Single Provider in DeepSeek-Reasonix Using the Models Array

You can group multiple model identifiers under a single provider entry in reasonix.toml by using the models = [...] array, allowing you to switch between them without duplicating credentials.

DeepSeek-Reasonix supports declaring several LLM variants behind one provider configuration. This approach lets you define base URLs and API keys once while exposing multiple model options to the UI. The feature is driven by the ProviderEntry struct in internal/config/config.go, which parses the TOML and populates a dropdown menu for model selection.

The ProviderEntry Struct Behind Multi-Model Support

The configuration logic resides in internal/config/config.go, specifically in the ProviderEntry struct (lines 1295–1304). This Go type defines how Reasonix interprets provider declarations from your TOML file.

Key fields for multi-model configuration include:

• Models []string – The array of model identifiers exposed under this provider.
• Default string – Specifies which model is selected automatically when the session starts.
• ContextWindow int – Sets a shared token budget applied to all models in the array unless overridden.

When Reasonix initializes, it reads the models slice and generates a dropdown in the UI. Switching models changes only the model parameter in API requests while reusing the same base_url and authentication credentials.

Configuring the Models Array in reasonix.toml

To set up multiple models, declare a provider block with the models field containing a list of strings. Each string represents a distinct model identifier recognized by the provider's API.

Basic Syntax

Define the provider kind, base URL, and list every model you want available:

[[providers]]
name        = "deepseek"
kind        = "openai"
base_url    = "https://api.deepseek.com"
models      = ["deepseek-v4-flash", "deepseek-v4-pro"]
api_key_env = "DEEPSEEK_API_KEY"

Setting a Default Model

Use the default field to pre-select a specific model when Reasonix loads. If omitted, Reasonix automatically uses the first entry in the models array.

[[providers]]
name        = "deepseek"
models      = ["deepseek-v4-flash", "deepseek-v4-pro"]
default     = "deepseek-v4-flash"

Shared Context Windows

The context_window field applies to all models in the array, ensuring consistent token limits across variants:

[[providers]]
name           = "deepseek"
models         = ["deepseek-v4-flash", "deepseek-v4-pro"]
context_window = 1_000_000

You can fine-tune individual models using model_overrides if specific variants require different limits.

How Reasonix Loads Multiple Models at Runtime

When the application starts, the TOML parser populates the ProviderEntry.Models slice. The UI layer then renders this list as a dropdown menu, allowing you to switch contexts without editing configuration files.

According to the source code in internal/config/config.go, the runtime behavior follows this flow:

1. Configuration Loading: The ProviderEntry struct is instantiated from reasonix.toml.
2. Model Population: The Models field is populated with the string array from the TOML file.
3. Credential Reuse: All models share the BaseURL and APIKeyEnv defined in the parent provider entry.
4. Default Selection: If default is set, that model ID is selected; otherwise, index 0 of the Models slice is used.

This architecture means you can add or remove model identifiers from the array, restart Reasonix, and immediately see the changes reflected in the model selector.

Complete Configuration Example

Here is a production-ready reasonix.toml excerpt demonstrating a DeepSeek provider with multiple models and a designated default:

[[providers]]
name        = "deepseek"
kind        = "openai"
base_url    = "https://api.deepseek.com"
chat_url    = "/v1/chat/completions"

# Expose multiple models under this single provider

models      = ["deepseek-v4-flash", "deepseek-v4-pro", "deepseek-v4-chat"]

# Select the flash model by default for speed

default     = "deepseek-v4-flash"

api_key_env = "DEEPSEEK_API_KEY"
context_window = 1_000_000

# Optional: Override context window for the Pro variant only

# model_overrides = { "deepseek-v4-pro" = { context_window = 2_000_000 } }

This configuration is based on the example found in reasonix.example.toml (lines 59–66), which demonstrates the recommended structure for multi-model setups.

Summary

• Declare multiple models using the models = [...] array inside a single [[providers]] block in reasonix.toml.
• Reuse credentials—all models in the array share the same base_url and api_key_env defined in the provider entry.
• Set a default with the default field, or let Reasonix automatically select the first model in the array.
• Reference implementation is found in internal/config/config.go within the ProviderEntry struct definition.

Frequently Asked Questions

What happens if I don't specify a default model?

If the default field is omitted, Reasonix automatically uses the first model listed in the models array as the default selection when the application initializes.

Can I override settings for individual models in the array?

Yes. While context_window applies to all models in the provider entry, you can define model_overrides in the TOML to specify per-model parameters, such as different token limits for specific variants.

How do I switch between models in the DeepSeek-Reasonix UI?

Once configured, the models array populates a dropdown menu in the Reasonix interface. Selecting a different option from this dropdown changes the active model for the session without requiring configuration file edits or application restarts.

Is the models array backward compatible with single-model configurations?

Yes. The models field is additive; if you have existing single-model configurations using the model field (the legacy string field), they continue to work. However, migrating to the models array provides greater flexibility for future expansion.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}