kforge

Understanding LLM Providers & Models in KForge

Last updated: 01/02/2026

This document explains how KForge thinks about AI providers, models, presets, and cost. It is intentionally high-level and provider-agnostic.

Provider policies, pricing, and available models change frequently. This document is a living reference and may be updated over time.


1. Providers vs Models (important distinction)

In KForge, providers and models are separate concepts.

KForge does not hardcode pricing, limits, or guarantees. You are always in control of which providers and models you configure.


2. Types of providers

Cloud-native providers

These companies:

Examples include OpenAI, Anthropic (Claude), Google (Gemini), and Mistral.

What this means:


OpenAI-compatible gateways

These are independent services that:

Examples include OpenRouter, DeepSeek, and Groq.

Important: “OpenAI-compatible” means same API shape, not same models, pricing, or guarantees.

Trade-offs:


Custom (bring-your-own endpoint)

The Custom provider allows you to connect:

Characteristics:

This option is intended for advanced users who already know what they are connecting to.


Local runtimes

These run entirely on your own machine.

Examples include Ollama and LM Studio.

What this means:

Local models are ideal for:


3. About models and cost

There is no permanently “free” model.

Most cloud providers charge per token:

Even models labeled as “free” may:

For this reason, KForge treats all cost information as advisory only.


4. Model labels and usage modes in KForge

KForge may display model labels using color and usage mode.

These labels are guidance, not enforcement.

Cost (color labels)

Usage modes

A model’s color indicates expected cost. Its usage mode indicates when it should be used.

KForge never:


5. About presets

Presets are curated default models provided by KForge.

They exist to:

Presets:

The current snapshot of presets is documented in:

Presets are defined in the app today but are expected to move to remote configuration in the future.


6. Duplicate models across providers (intentional)

You may see the same model name appear under multiple providers.

This is intentional.

Examples:

These options differ in:

The capability may be similar, but the tradeoffs are not.


7. Volatility and expectations

Some models are inherently volatile:

This is normal.

KForge is designed so that:

Volatility is not a bug — it is part of the ecosystem.


8. Choosing a model (practical advice)

Instead of asking “what is the best model?”, consider:

You can switch models at any time. There is no single correct choice.


9. Why this document lives outside the app

Provider models, pricing, and policies change constantly.

Embedding this information directly in the app would:

By linking to documentation instead, KForge stays:


10. Core principle

KForge does not sell models.
KForge does not guarantee cost.
KForge gives you control.