Set up your AI provider

1. Open Google AI Studio → API keys and sign in with a Google account.
2. Choose Create API key, and pick or create a Google Cloud project for it.
3. Copy the key and paste it into API key.
4. Model: gemini-3.8-flash (balanced) or gemini-3.5-flash-lite (cheapest).

The free tier has daily request limits; RepoLantern pauses when one is reached and resumes after the reset. Read Google's terms for the free tier before sending private code: use a key on a paid (billing-enabled) project for that.

1. In the Google Cloud console, choose a project with billing, and enable the Vertex AI API.
2. Preferred: an API key. Create one for Vertex AI in your project and restrict it to the Vertex AI API. Paste it into API key.
3. Or a service-account key: in IAM → Service accounts, create an account just for RepoLantern with only the role Vertex AI User (roles/aiplatform.user). Under Keys, add a JSON key, open the file, and paste its contents into Use a service-account key instead. Delete the key when you're done: it is long-lived.
4. Location: leave empty for global, or type a region such as us-central1.
5. Model: gemini-3.8-flash or gemini-3.5-flash-lite.

1. Open the Claude Console → API keys and sign in.
2. Choose Create key, name it "RepoLantern", and pick a workspace. A workspace with its own spend limit keeps costs contained.
3. Copy the key (it starts with sk-ant-) and paste it into API key.
4. Model: claude-haiku-5-5 (cheapest), claude-sonnet-5-5 (balanced) or claude-opus-5-5 (most capable).

1. Open OpenAI → API keys and sign in.
2. Choose Create new secret key, name it "RepoLantern", and pick a project. Set a budget on that project.
3. Copy the key (it starts with sk-) and paste it into API key. It is shown only once.
4. Model: gpt-6-luna (recommended) or gpt-5.4-mini.

Choose the service in Service and the endpoint fills itself in. The model name comes from that service's model list. For OpenRouter, RepoLantern uses the price OpenRouter publishes for that exact model (shown as OpenRouter's published price, with a link to the model's page). For the other services it doesn't know the prices, so enter yours under Advanced (per million tokens) to keep the estimate and limit right.

Any other HTTPS service that speaks the OpenAI Chat Completions API works too: choose Another OpenAI-compatible service and type its endpoint. Endpoints must be public https:// addresses.

Free models on OpenRouter

• What ":free" means. A model id that ends in :free (for example nvidia/nemotron-3.5-super-vl:free) is a free version of that model with its own, lower limits. RepoLantern shows its price as $0, from OpenRouter's own list.
• How to pick one. On openrouter.ai/models, look for models whose price says free, and copy the id that ends in :free into Model. Choose one that handles long text well.
• Limits. OpenRouter allows 20 requests a minute for free models, and 50 a day until your account has bought 10 credits (then 1,000 a day) (OpenRouter's limits). The daily number counts all your free-model requests together. RepoLantern paces itself to 20 a minute and pauses at 50 a day; if your account has more, raise the daily number under Advanced → Your model's limits. The estimate shows how many requests a guide needs.
• Your code and privacy. Your code goes to OpenRouter and to the company that runs the model. OpenRouter says it doesn't store prompts unless you opt in (data collection). The company behind each model has its own policy, and some may keep prompts or train on them. Your OpenRouter privacy settings decide whether those may be used, with a separate setting for free models (provider logging). Check them before you send private code.
• Free models can go away. OpenRouter can withdraw a free model; its list gives the date when one is planned, and RepoLantern shows that date next to the price. If a model is gone, pick another one.
• Your spending limit still applies. A free model costs $0, so the limit matters only for a paid backup model.

Ollama

• Ollama Cloud works today: create a key at ollama.com/settings/keys, choose Ollama Cloud, and use a model name from ollama.com/api/tags (for example gemma4:31b). Ollama Cloud is a plan rather than a per-token price: if your plan is flat-rate, enter 0 as your price under Advanced.
• Ollama on your own computer isn't available on the hosted RepoLantern: to our server, "localhost" is the server itself, and private addresses are blocked on purpose.

With a Bedrock API key (recommended):

1. In the Bedrock console, pick your region and check the models you want are available to your account.
2. Open API keys in the Bedrock console and generate a key. A short-term key (it expires with your console session) is safest; a long-term key lets you set an expiry. Copy it when it is shown.
3. On the Model step choose AWS Bedrock, paste it into Bedrock API key, and choose the same Region. RepoLantern accepts these regions: us-east-1, us-east-2, us-west-2, ca-central-1, sa-east-1, eu-west-1, eu-west-3, eu-central-1, ap-south-1, ap-northeast-1, ap-southeast-1 and ap-southeast-2.
4. Model: the model ID, or for newer models an inference-profile ID (these start with us., eu. or apac.), for example us.amazon.nova-micro-v1:0.

A new AWS account may answer "Your account is currently being verified" or "Operation not allowed" until AWS finishes verifying it: RepoLantern pauses and says so; try again once AWS confirms the account.

With an access key pair instead:

1. In IAM, create a user just for RepoLantern (no console access) with an inline policy that allows only bedrock:InvokeModel, bedrock:ListFoundationModels, bedrock:ListInferenceProfiles and bedrock:GetInferenceProfile.
2. On that user's Security credentials tab, create an access key ("Application running outside AWS").
3. On the Model step open Use an access key pair instead and paste both values. The session token is only for temporary credentials.
4. Delete the access key when you're done.

1. Open Microsoft Foundry and open (or create) a project.
2. Deploy a model and note the deployment name you gave it: that is the model name in RepoLantern.
3. From the resource's Keys and Endpoint, copy the endpoint (https://<resource>.openai.azure.com or https://<resource>.services.ai.azure.com) and one of the keys.
4. Paste them into Endpoint URL and API key, and type the deployment name as the model.

RepoLantern can search the web for what changed in newer versions of the libraries a repository uses. It needs a Tavily key: sign up, copy the key (it starts with tvly-) from the overview page, and paste it under Settings → Web search or on the last step of New guide, then press Test. The free plan gives 1,000 credits a month.

Give a read-only token that can see only the repositories you choose. It is used once to fetch the code and never stored.

• GitHub: Settings → Developer settings → Personal access tokens → Fine-grained tokens → Generate. Under Repository access choose Only select repositories; under permissions set Contents: Read-only. Pick a short expiry.
• GitLab: a project or personal access token with only the read_repository scope.
• Bitbucket: a repository access token (or API token) with Repositories: Read only.