Setting Up API Providers

ThinkForge connects to external AI model providers through API keys, so you choose which service powers chat, agents, skills, and other AI-assisted workflows.

An API provider is the service ThinkForge sends AI requests to. Examples include OpenAI, Anthropic, Gemini, and OpenRouter. OpenRouter is especially useful because it gives access to many model families through one OpenAI-compatible API endpoint.

Why API providers matter

Different models are better at different work. One model may be stronger for coding, another for writing, another for fast summaries, and another for research. ThinkForge is built so your workspace, projects, skills, and agents do not depend on a single provider.

With API provider settings, you can:

  • Bring your own API key.
  • Choose a default provider and model.
  • Use OpenRouter to access many model families from one account.
  • Change models as your needs change.
  • Set a specific provider and model for individual agents or AI interfaces.
Your ThinkForge workspace stays yours. The provider is the model layer you choose to connect.

Supported providers

ThinkForge currently includes provider definitions for:

  • OpenAI
  • OpenRouter
  • Anthropic
  • Gemini

Each provider has a default endpoint and default model. OpenRouter uses an OpenAI-compatible endpoint:

https://openrouter.ai/api/v1

When you use OpenRouter, you can select the model provider family separately, such as OpenAI, Anthropic, Gemini, DeepSeek, Qwen, xAI/Grok, Mistral, or Llama, then choose a model from that provider family.

Before you start

You need an API key from the provider you want to use. For example:

  • OpenAI keys come from your OpenAI account.
  • Anthropic keys come from your Anthropic account.
  • Gemini keys come from your Google AI/Gemini setup.
  • OpenRouter keys come from your OpenRouter account.
Privacy note: ThinkForge stores saved keys locally using its API key store. The key is used when ThinkForge sends requests to the provider you selected.

Set the global default provider

The global default is the provider and model ThinkForge uses when a specific tool, chat surface, or agent has not been given its own override.

To set it up:

  1. Open ThinkForge.
  2. Go to Settings.
  3. Open API Key or AI Routing Settings.
  4. Select the Global Default profile.
  5. Choose a provider.
  6. Choose a model.
  7. Paste the API key for that provider.
  8. Adjust optional settings such as temperature, max tokens, or endpoint if needed.
  9. Save the default profile.

After this is saved, ThinkForge can use that provider anywhere no more specific provider has been selected.

Using OpenRouter

OpenRouter works a little differently from a direct provider because it acts as a routing layer for many model providers.

When setting up OpenRouter:

  1. Choose OpenRouter as the provider.
  2. Paste your OpenRouter API key.
  3. Leave the endpoint as the default unless you have a specific reason to override it.
  4. Choose the model provider family.
  5. Choose the model you want to use.
  6. Save the profile.

OpenRouter model names often include the model family in the name, such as:

openai/gpt-4o-mini
anthropic/claude-sonnet-4.6
google/gemini-flash

This makes OpenRouter a good choice if you want one key that can reach several model families.

Set a provider for a specific agent

ThinkForge also supports provider and model selection at the agent or interface level.

This means you can keep a global default for general use, then give a specific agent a different provider or model when that agent needs different strengths. For example:

  • A writing agent can use a model that is strong at tone and structure.
  • A coding agent can use a model that is strong at code reasoning.
  • A research agent can use a model with a larger context window.
  • A fast utility agent can use a cheaper, faster model.

In agent or AI model settings, you can choose whether to use the global default or override it. When Use Default Profile is enabled, the agent follows the global setting. When it is disabled, you can choose a provider and model just for that agent.

This keeps routing flexible without forcing every workflow to use the same model.

When to use the global default

Use the global default when you want simple setup and consistent behavior. This is best when:

  • You are just getting started.
  • You mainly use one provider.
  • You want every AI surface to behave the same way.
  • You do not want to manage separate model choices yet.

When to use agent-specific providers

Use per-agent provider settings when the agent has a specialized job. This is best when:

  • One agent needs stronger reasoning.
  • One agent should be faster or cheaper.
  • One workflow needs a specific model family.
  • You want to test multiple models without changing the whole app.
  • You are building agents for different types of work.

The global default gives you a baseline. Agent-specific settings give you control.

Endpoint overrides

Most users should leave the endpoint field unchanged. ThinkForge already knows the standard endpoint for each built-in provider.

Endpoint overrides are useful when:

  • You are using a compatible proxy.
  • Your organization has a custom gateway.
  • You are connecting to a provider-compatible endpoint.
  • You need to test a nonstandard routing setup.

If you are not sure, keep the default endpoint.

Troubleshooting

If requests fail after setup, check the basics first:

  • The API key is pasted correctly.
  • The key belongs to the provider you selected.
  • The provider account has billing or credits enabled if required.
  • The selected model is available to your provider account.
  • The endpoint was not changed accidentally.
  • For OpenRouter, the selected model exists and is available through OpenRouter.

If one agent fails but other AI features work, check whether that agent is using its own provider override.

The main idea

ThinkForge separates your workspace from your AI provider. Your projects, files, memory, and agents stay in ThinkForge. The model provider is a configurable layer.

That means you can start simple with one default provider, then route specific agents to the best model for their job as your workflows grow.

Related