Desktop model setup

This page covers desktop model configuration. To use the KeenKind API in other coding tools, see the API quickstart and Tool setup.

KeenKind supports popular providers, custom compatible APIs, and local services. Model configuration and keys are stored locally.

Add a configuration

1. Open Settings > Configuration.
2. Click Add configuration.
3. Choose a provider and enter its endpoint and API key.
4. Add at least one model ID.
5. Save and enable the configuration.

Known providers use their corresponding API type. Do not select OpenAI simply because a service is OpenAI-compatible; for example, select DeepSeek for DeepSeek.

Custom APIs

If a service is not listed, choose a custom configuration. Confirm which API it implements:

• OpenAI Responses
• OpenAI Chat Completions
• Anthropic Messages
• Gemini

The API type must match the service’s implementation. Working chat alone does not establish support for tools, images, thinking content, or streaming.

Local models

KeenKind includes common endpoint templates for Ollama, vLLM, and LM Studio. Local services usually do not need an API key, but the service must be running with a model loaded.

Model options

Set each model’s display name, ID, context window, maximum output, and reasoning level. Use defaults when provider limits are unknown.

For image models supporting generation or editing, you can also set:

• Support for generation, editing, and transparent backgrounds.
• Maximum outputs per request (up to 8) and reference images per edit (up to 5).
• Supported sizes, quality, formats (PNG, JPEG, WebP), and their defaults.

If image capabilities are unclear and generation fails, try fewer output or reference images.

KeenKind remembers the most recently used configuration and model. It chooses another available model only if the previous configuration is disabled, deleted, or missing the model.

Account entry

An available custom API configuration hides the account and sign-in entries. You do not need a KeenKind account to use your own API.

Troubleshooting

Symptom	Check
401 or 403	API key, account permissions, and request endpoint.
404	Base URL, version path, and model ID.
Chat works but tools fail	Model and API tool-calling support.
Empty response	Streaming format, maximum output, and model capabilities.
Thinking content is missing	The model provides it and display is enabled in settings.

Caching is controlled by each provider. Cache hit rates can temporarily fall on first use or after changing model or mode.