kforge

Custom Provider (OpenAI-compatible Endpoints)

Last updated: 28/01/2026

---

What is the Custom provider?

The Custom provider lets you connect any OpenAI-compatible API endpoint to KForge.

If your provider supports:

• POST /v1/chat/completions
• Authorization: Bearer <API_KEY>

Then it will work with Custom.

This makes KForge future-proof and vendor-agnostic:

• You are not locked into specific providers
• You can use new providers immediately
• KForge keeps working even if providers change, rotate, or disappear

Custom is designed for:

• advanced users
• enterprise setups
• self-hosted gateways
• anyone who wants full control

---

What Custom is NOT

Custom does not:

• fetch models automatically
• know pricing, quotas, or rate limits
• guarantee availability of any model
• provide “free” models by itself

All of this depends entirely on:

• your endpoint
• your provider account

KForge is honest about this by design.

---

Endpoint requirements

Your endpoint must be OpenAI-compatible.

Required

• Base URL only (no trailing /v1)
  • Example: https://api.mistral.ai
• Authentication via Bearer token

KForge will automatically call:

• /v1/chat/completions

If you enter a URL ending in /v1, KForge will normalize it for you.

---

How to configure Custom in KForge

1. Select Custom Endpoint (OpenAI-compatible) as the provider
2. Open Settings
3. Enter:
   • API Key: your provider’s API key
   • Endpoint URL: base URL (no /v1)
4. Save settings
5. Add or select a model ID supported by your endpoint

That’s it.

---

Suggested models (examples only)

These are suggestions, not presets.

Availability, pricing, and limits may change at any time depending on the provider. Use them as a starting point, then adjust based on your usage.

---

OpenRouter

Endpoint:

• https://openrouter.ai/api

Example models:

• llama-3.1-8b-instant — Free / Sandbox
• openai/gpt-4o-mini — Paid / Main

Notes:

• Free models may rotate or be deprecated
• Rate limits can apply

---

Groq

Endpoint:

• https://api.groq.com/openai

Example models:

• llama-3.1-8b-instant — Free
• llama-3.3-70b-versatile — Main

Notes:

• Extremely fast inference
• Pricing depends on your Groq account

---

DeepSeek

Endpoint:

• https://api.deepseek.com

Example models:

• deepseek-chat — Free
• deepseek-reasoner — Main

Notes:

• Free tier may have limits
• Reasoner is slower but stronger

---

Mistral

Endpoint:

• https://api.mistral.ai

Example models:

• mistral-small-latest — Sandbox / Main
• codestral-latest — Main (coding-focused)

Notes:

• Requires API key
• Free evaluation tiers may be limited

---

Model tags and usage modes (important)

Model tags in KForge are organizational hints, not guarantees.

They reflect how you intend to use the model — not what the provider promises.

Suggested interpretation

• Sandbox
  • Cheap or free models
  • Safe for iteration and experimentation
• Free
  • No billing required
  • Often rate-limited or restricted
• Main
  • Daily-driver models
  • Reasonable cost and capability
• Heavy
  • Expensive or high-capability models
  • Use sparingly
• Unknown
  • Pricing or limits unclear
  • Check provider documentation

For Custom endpoints, you are the source of truth.

---

Why Custom matters

Custom exists so that:

• You are never blocked by missing integrations
• You can use private, enterprise, or self-hosted endpoints
• KForge survives provider churn
• Advanced users are not limited by UI assumptions

Many tools hard-code providers. KForge deliberately does not.

---

When should I use Custom?

Use Custom if:

• your provider is not listed in KForge
• you run your own OpenAI-compatible server (vLLM, TGI, gateways)
• you want full control over models and endpoints
• you understand your provider’s pricing and limits

If you want “plug-and-play”, use a built-in provider instead.

---

Final note

Custom is powerful by design.

With power comes responsibility — and flexibility.

That trade-off is intentional.