Frequently asked questions

Short answers to the questions people ask most.

Questions about RepoLantern

What does RepoLantern do?

You paste a Git repository link and choose an AI model with your own key. RepoLantern reads the whole codebase and writes a complete guide for beginners, senior engineers and architects, with architecture diagrams, setup steps, a glossary and citations to the exact lines of code.

Is RepoLantern free?

Yes. RepoLantern itself is free. You bring your own model key (Gemini, Vertex AI, Claude, OpenAI, AWS Bedrock, Microsoft Foundry, or an OpenAI-compatible service such as OpenRouter, Groq or Ollama Cloud), so you pay only your model provider, and you see the likely cost before anything is spent. Free models on OpenRouter are shown at $0.

Is my API key stored?

No. Your key stays in your browser tab's memory and is sent only with the request that needs it. It is never written to a database, a log or a file, and it is gone when you close the tab.

How does RepoLantern avoid hallucinations?

Every file path, function and class name in a guide is checked against the real code. Sections with names that can't be found are rewritten once and kept only if they improve; anything still unverified is listed under "Names we couldn't find in the code". Each guide gets a grounding score, and citations link to the exact lines at the analysed commit.

Can it document private repositories?

Yes, on any HTTPS Git host (GitHub, GitLab, Bitbucket or your own server). You give a read-only access token for that one run; it is used once and never stored. Secrets found in the code are removed before anything is sent to a model.

Which formats can I export?

Markdown, PDF, Word (DOCX), PowerPoint (PPTX), HTML and editable draw.io diagrams, in your own brand colours.

Who is RepoLantern for?

Developers joining an unfamiliar codebase, senior engineers and architects reviewing a system, teams documenting code before a handover, consultants assessing a client's code, and anyone building agentic AI applications who wants to understand an open-source project quickly.

How long does a guide take?

Small repositories take a minute or two; large ones longer. The estimate shows the expected time, and the job keeps running on the server if you close the tab.

What is AELSHA™ ASTRA?

AELSHA™ ASTRA is AELSHA™'s engineering framework for taking AI agents to production. RepoLantern is built with it, so every guide goes through gated steps (you approve the cost, the guide is checked against the code before you see it), every model and tool call is traced and costed, checks are written in code and tested on every release, and spending, consent and the choice of model stay with you. See https://repolantern.aelsha.com/astra.

How is RepoLantern different from asking a chat assistant about my code?

RepoLantern reads the whole repository at one exact commit instead of the few files you paste into a chat. It then checks every file path, function and class name against the code, lists the names it could not verify, and gives the guide a grounding score.

What is a grounding score?

A grounding score is a number from 0 to 100 that says how well a guide sticks to what is really in the code. It counts whether the files, functions and classes the guide names exist, whether its citations resolve and whether its required sections are present; it does not judge whether every explanation is right.

Which programming languages does it support?

RepoLantern recognises more than 40 languages and file types and can document a repository in any of them. Names are checked exactly for Python and by pattern matching for JavaScript, TypeScript, Vue, Svelte, Go, Java, Kotlin, C#, Rust, Ruby, PHP and Swift; for other languages the guide is still written, but name checking is limited.

How accurate is it?

In a blind round written by a separate reviewer, the checks caught 94.3% of planted invented names, and 83.3% in a second, harder round; real names were never flagged as invented. All 12 golden guides reproduce exactly. The method and results are at https://repolantern.aelsha.com/evals.

How do I understand a new codebase quickly?

Run it, find the entry points, trace one request and read the tests, then check every claim against the code. RepoLantern does the first pass for you. Paste the repository link and get a guide with architecture diagrams, a quick start and a glossary, with citations to the exact lines. The step-by-step method is at https://repolantern.aelsha.com/learn.

Can I generate documentation or an architecture diagram from a GitHub repository?

Yes. Paste the repository link and RepoLantern writes a guide for the depth you choose, with architecture diagrams drawn from the code's real imports. Export the guide to Markdown, PDF, Word, PowerPoint or HTML, and the diagrams as images or editable draw.io files.

Can I use my own API key, and which models work?

Yes, RepoLantern only works with your own key. It supports Google Gemini, Vertex AI, Anthropic Claude, OpenAI, AWS Bedrock, Microsoft Foundry and OpenAI-compatible services such as OpenRouter. The key is never stored, and the likely cost is shown before anything is spent.

Does it help with an agentic AI or LangGraph project?

Yes. It documents any repository, and the pipeline treats agent code like other code, so names and citations are checked against the source. Two articles explain what to look for by hand, in agentic AI codebases and in LangGraph projects, at https://repolantern.aelsha.com/learn.