Bring Your Own Keys

ThinkForge uses your own AI provider API keys. You choose which providers to use, configure your keys, and control which model handles each part of the app.

ThinkForge does not bundle or resell AI access. You sign up with providers directly, get your own API keys, and enter them in ThinkForge. This means you control costs, privacy policies, and model selection.

Supported providers

  • OpenAI — GPT-4o, GPT-4, and other OpenAI models
  • Anthropic — Claude 3.5, Claude 3, and other Anthropic models
  • Google Gemini — Gemini 2.0 Flash and other Gemini models
  • OpenRouter — access multiple providers through a single API key, including models from OpenAI, Anthropic, Meta, Mistral, and others

Custom endpoints are also supported for OpenAI-compatible APIs, local model servers, or proxy gateways.

Setting up API keys

  1. Open Settings → AI Routing Settings in ThinkForge.
  2. Select a provider.
  3. Enter your API key. You can store multiple keys per provider if needed (for example, separate keys for different billing accounts).
  4. Choose a default model for that provider.
  5. Save.

Once a key is configured and valid, ThinkForge can refresh the available model list from the provider's API, showing you exactly which models your key has access to.

AI profiles

Each provider configuration forms an AI profile with settings for:

  • Provider and API key
  • Model selection
  • Temperature (creativity vs. consistency)
  • Max tokens (response length limit)
  • Optional endpoint override (for custom or local deployments)

Per-interface routing

ThinkForge lets you route different parts of the app to different providers and models. You set a global default, then optionally override it for specific interfaces:

  • Desktop conversation workspace — conversations inside ThinkForge Desktop
  • Agent creation — the Meta Agent planner that designs agent pipelines
  • Agent execution — individual agent deployments
  • Planning and orchestration — internal planning tasks

For example, you might use a fast, affordable model for chat and a more capable model for agent creation. Each interface shows a model selector when multiple options are available.

What uses your keys

Your API keys are used for:

  • Desktop conversation workspace
  • Agent AI steps (processing, summarizing, transforming content)
  • Agent creation (the planner that designs pipelines)
  • Any feature that calls an AI model for text generation

Your API keys are not used for:

  • Embedding generation — this uses a local model (no API call)
  • Search and indexing — runs entirely on your machine
  • File management, project organization, or UI operations
Keys stored locally. Your API keys are stored in a local database on your machine. They are not sent to ThinkForge servers, shared with other users, or used for any purpose other than calling the provider you configured.

Cost control

Since you use your own API keys, you pay providers directly at their published rates. ThinkForge does not add markup. To manage costs:

  • Choose smaller or faster models for routine tasks
  • Use the per-interface routing to assign expensive models only where they matter most
  • Set max token limits to prevent unexpectedly long responses
  • Monitor usage through your provider's dashboard

Related