Getting started

This page takes you from a fresh account to an agent answering questions about your code with real citations. Budget about ten minutes, most of which is waiting for summaries to generate.

You will need one thing we cannot give you: an API key from an LLM provider. BNLGit never bills you for model usage and never uses a key of ours to generate your summaries — see BYOK connections.

1. Sign in

Your trial signup creates your forge account and emails you a link to set your password. Once that is done you sign in at app.bnlgit.com.

Self-service registration on the forge is switched off. Accounts are created only by the signup flow, which means there is no open registration surface for strangers to abuse.

2. Save a provider key

Go to Settings → Connections & BYOK and add an LLM connection. Pick your provider, paste your key, and give the connection a short reference name — you will use that name when you pair a repository.

Then press Verify, and then Refresh models. Refreshing pulls the model catalogue from your provider, and BNLGit will only let you select a model that catalogue actually returned. There is no free-text model box, because a typo'd model ID that fails at generation time three hours later is a worse experience than being told now.

Supported LLM providers: litellm, openrouter, fireworks, anthropic, azure_openai, and any openai_compatible endpoint.

3. Get a repository onto the forge

Either push an existing one:

git remote add bnlgit git@app.bnlgit.com:your-org/your-repo.git
git push bnlgit --all
git push bnlgit --tags

…or create a new repository in the UI. Nothing about this step is BNLGit-specific — it is ordinary Gitea. If you are migrating from another host, see Coming from Gitea.

4. Pair it

From the repository page, use the BNLGit pairing action, or call the API directly:

POST /api/v1/repos/{owner}/{repo}/bnlgit/pair
Authorization: token <your token>
Content-Type: application/json

{
  "llm_credential_ref": "openrouter-primary",
  "llm_model": "anthropic/claude-sonnet-5"
}

Both fields are required and explicit. BNLGit will not infer which of your connections to spend, and will not pick a default model on your behalf.

Pairing creates your-repo-llm beside your source repository. It does not generate anything yet, and it does not spend a cent.

5. Generate summaries

Open the BNLRepo Sync Cockpit at /{owner}/{repo}/bnlgit. You will see every eligible file and its state — pending for everything, at this point.

Press Generate. You will be shown a cost estimate and asked to confirm it before any billable call is made. That confirmation is enforced on the server, not just in the browser: a bare API POST cannot queue billed work without it. See Managed worker and spend controls.

Summaries land in the mirror repo as files. You can git clone your-repo-llm and read them in your editor, diff them, or grep them — they are just text in a Git repository.

6. Point an agent at it

This is the step where the product earns its keep. Full detail is in Connect your agent (MCP); the short version is that you mint a normal Gitea API token and hand your agent this URL:

https://app.bnlgit.com/api/v1/repos/{owner}/{repo}/bnlgit/mcp

Ask it something you would otherwise have to read the code to answer — "where do we validate the session token, and what happens when it has expired?" — and check the citation it gives you against the real file. That check is the whole point: you should be able to verify the answer in one click.

What to do next

• Read The BNLRepo mirror model to understand freshness states and the drift signal, so you know when to trust a summary.
• Read Search before you rely on semantic search — there is an important limitation documented there.
• Read Managed worker and spend controls before you turn on background generation for a large repository.