Quickstart — connect your coding agent

Connect your local coding agent to Dependra. Install the CLI, sign in once, and your agent can take inventory of a repository, push it into your landscape, and answer questions about impact and risk — all attributed to you.

1. Install the CLI

npm install -g dependra      # or run any command with: npx dependra <command>

Then, inside a repository, wire your agent and scaffold a starter manifest:

dependra init

2. Sign in

dependra login

This opens your browser; you approve, and the CLI stores a short-lived token. There are no API keys to manage for local work, and everything your agent and dependra push do is attributed to you in the audit log. Re-run dependra login to renew.

Once you’re signed in, your agent can also query your live landscape while it works (read-only) — so it matches components that already exist instead of creating duplicates.

3. Take inventory of a repo and push it

With your agent connected, ask it to take inventory:

“Take inventory of this repository and author the Dependra manifest.”

Your agent reads the codebase and writes .dependra/manifest.yaml — the applications, components, dependencies and datasets this repo implements. Review it, then push:

dependra push --preview   # see exactly what would change — nothing is written
dependra push             # apply it to your landscape

--preview is a dry run (think git push --dry-run); a bare push applies the changes you approve.

4. Keep it in sync from CI

Your landscape should move at the speed of your code. Wire the push into your pipeline so every merge keeps Dependra current.

GitHub Actions — scaffold the workflow:

dependra github

This drops in a ready-to-use workflow that authenticates with GitHub’s built-in OIDC — no secrets to store. Approve the repository once in your Dependra organisation settings, and every push keeps your landscape current.

GitLab or any other CI — authenticate with a scoped API key instead. Create a key in your Dependra organisation settings, expose it to the job as DEPENDRA_API_KEY, and run npx dependra push as a pipeline step.

5. Ask your architecture

Because your agent is connected to Dependra, you can just ask — in plain language — and it queries your live landscape to answer:

• “What component in my landscape has the most concentrated risk?”
• “What breaks if the payments queue goes down, and who’s affected?”
• “Which applications depend on the orders database?”

Answering is read-only — asking never changes your model. Every change still flows through dependra push, which you review and approve.