LogoFastRouter
Back
Your Fine-Tuned Models Now Work Inside FastRouter

Your Fine-Tuned Models Now Work Inside FastRouter

Add fine-tuned and custom model endpoints to FastRouter. Route them like any standard model — with full observability, cost tracking, and governance.

Andrej Gamser
Andrej Gamser
3 Min Read|Latest -

Bring Your Own Key (BYOK) on FastRouter has always let you attach your own provider API keys. We've extended it: you can now add custom model endpoints — fine-tuned models, self-hosted models on supported providers — and route them through FastRouter alongside every other model in your stack.

The Problem

Teams that fine-tune models or host custom models on providers like OpenAI, Azure, or AWS Bedrock end up managing two separate integration layers: their custom model endpoint and whatever gateway they use for standard models. The result is split observability, split cost tracking, and split governance — a custom model that doesn't benefit from any of the routing, fallback, or monitoring infrastructure the rest of the stack relies on.

What BYOK Extended Lets You Do

BYOK on FastRouter lets Organisation Owners attach external provider API keys and custom model configurations directly to their organisation. Once added, those models are routed through FastRouter like any other model — benefiting from the same observability dashboard, rate limit controls, access governance, and cost tracking.

Supported providers for BYOK:

Anthropic, OpenAI, Google AI Studio, Google Vertex AI, AWS Bedrock, Azure, Grok, Groq, DeepInfra, Mistral, Moonshot AI, Baseten, Fal.AI, Pollo.AI, Perplexity

When using BYOK, rate limits and costs are determined by your own provider account, not by FastRouter.

Adding a Custom Azure Model Endpoint

Azure BYOK supports an array of configurations, allowing you to register multiple custom model endpoints in a single setup. Each entry maps a FastRouter model slug to a specific Azure deployment:

json

1[
2  {
3    "model_slug": "openai/gpt-5.4",
4    "endpoint_url": "YOUR-AZURE-ENDPOINT-URL",
5    "api_key": "YOUR-AZURE-API-KEY",
6    "model_id": "gpt-5.4"
7  }
8]

Make sure model_slug matches the FastRouter model name and model_id aligns with the model name configured in your Azure account.

Adding AWS Bedrock

AWS Bedrock BYOK uses a single configuration object containing your access credentials and the target AWS region:

Ensure your credentials have the necessary permissions for invoking Bedrock models in the target region.

Who Can Add External Keys

Only Organisation Owners can add, update, or remove provider keys and set provider-specific preferences. This ensures administrative control over billing, usage, and access.

Why This Matters

If you've fine-tuned a model — a domain-specific classifier, a task-optimised assistant, a cost-reduced distilled model — you can now route it through FastRouter. That means:

  • Your fine-tuned model appears in the same observability dashboard as every other model
  • You can set up fallback chains that route to your custom model first and fall back to a standard model if it fails
  • Cost tracking and governance apply to custom model calls the same as to standard provider calls
  • You can use Dynamic Tags to route specific workloads to your custom model without changing application code

Get Started

BYOK is available to Organisation Owners now. Navigate to Add External Keys in your FastRouter dashboard to get started.

Documentation: docs.fastrouter.ai/add-external-keys-byok

FastRouter — The LLM Gateway Built for Production | fastrouter.ai

Related Articles