Use your own AI API key for research appraisal
Connect OpenAI, Claude or Gemini directly, choose a model, and understand what MedLit sends and stores.
By the MedLit team · Updated October 13, 2026
Direct provider connections
MedLit supports the OpenAI API, Anthropic Claude API, Gemini Developer API and optional OpenRouter API. These are distinct services. Choose the provider that issued your key. Azure OpenAI, Vertex AI service accounts, AWS Bedrock credentials and Claude Code login tokens use different authentication contracts and are not accepted as interchangeable API keys.
Choose and test a text model
The connection dialog can load the provider’s model catalog. You can also enter a model ID, which is useful when a restricted key permits generation but not catalog access. Test and connect sends a brief prompt without a paper to check generation access. This small test and later appraisals may incur charges in your provider account.
Understand the privacy boundary
On the website, the API key stays in page memory until refresh or disconnect. It is sent directly to the selected provider, not stored in the MedLit database or analytics. An appraisal sends the retrieved article and review instructions to that provider. Use material you are permitted to share and consider the provider’s data settings.
Accounts and subscriptions are different connections
MedLit account access and AI credits are separate. Library cloud sync requires an activated cloud account; device saving is available without one. Standard API keys use provider API billing. OpenAI now documents a limited commercial Sign in with ChatGPT program, but MedLit has not activated that integration. A manual source packet can instead be used in ChatGPT, Claude or Gemini and imported back for evidence checking.
Try it with a paper
Explore the teaching example without an account, or bring a paper and connect your own AI provider.
Explore the example appraisalAppraise a paper in MedLit