DL
DevLoops

Get started with DevLoops.

A step-by-step guide. Take your time — none of this is a race.

First time here? You're going to set up Redis (we handle that), download a CLI, run one command, and start asking questions about your codebase. Total time: commit ingestion takes approximately 2 minutes per commit. The front end CLI tool is processing commits really quickly so you dont have to keep your laptop open but our server can work overnight if needed! We'll tell you what to expect at every step.

Setup — five steps

1
Sign in to your dashboard
~30 seconds

If you haven't already created an account, sign up. Otherwise sign in. The dashboard is where you'll provision Redis, download the CLI, generate API keys, and watch ingestion progress.

Sign up → Sign in →
You're done with this step when: you can see the DevLoops dashboard with your name in the header.
2
Provision your Redis Cloud database
~3 minutes

DevLoops stores your codebase intelligence in a private Redis Cloud database — one per organization, fully isolated. From the dashboard, scroll to the REDIS SETUP section, pick a plan (the free tier works fine to start), and click "Provision Redis."

We handle the entire setup. You don't need a Redis account. You won't see a credit card prompt unless you choose a paid tier.

Open dashboard →
You're done with this step when: the dashboard shows "Connected" with a green dot in the INDEX section, and the Redis setup section disappears.
3
Download the DevLoops CLI
~1 minute

The CLI is what walks your git history and ships symbol metadata to our backend. Download it from the dashboard's CLI section. It's a single shell script — no npm install, no Python venv, no Docker.

After downloading, make it executable:

chmod +x devloops.sh
Download CLI →
You're done with this step when: running ./devloops.sh --help in your terminal shows usage instructions.
4
Ingest your repo
5 minutes to several hours

Grab your JWT token from the dashboard (the API ACCESS section, "Your JWT token" — click show then copy). Then run:

./devloops.sh bootstrap \
  --repo /path/to/your/repo \
  --token YOUR_JWT_TOKEN \
  --commits 50

Start with --commits 50 for a small sample so you can see things working before committing to a full ingestion. Once you're confident, drop the flag to ingest everything.

Big repos can take hours. The CLI shows progress per commit. You can close your terminal and the dashboard shows a live progress bar — server-side processing keeps going.

You're done with this step when: the dashboard's "Active skills" count is greater than zero. Refresh the page to see the latest count.
5
Ask your first question
~10 seconds

Open the chat interface and ask anything about your code. Try:

How does authentication work in this codebase?
What are the main entry points?
Walk me through how X works.

Pick Standard mode for most questions (~$0.10 per query). Use Quick for fast lookups, Deep for security reviews and cross-cutting analyses.

Open chat →
You're done with this step when: you've asked a question and gotten an answer that references real symbols from your code. That's it — you're up and running.
Stuck somewhere? Email a.anapolitanos@gmail.com with what step you're on and what you're seeing. We're a small team — we read every message and reply within a day.

Want to integrate DevLoops into your own agent?

+
Generate an API key and call /api/v1/query
~2 minutes

From the dashboard's API KEYS section, click "Generate new API key." Save the key immediately — it's shown only once.

Then call the API from anywhere:

curl -X POST https://www.devloops.ai/api/v1/query \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "How does the auth flow work?",
    "mode": "standard"
  }'

Returns JSON with the answer and the source skills used to generate it. Use this to integrate DevLoops into Claude Code, Cursor, custom agent workflows, or anything else that can make HTTP requests.

Generate API key →

The bigger picture

DevLoops watches your git commits, extracts the structure of your code (functions, classes, mechanisms), describes each one in natural language, and stores the result as a queryable graph in your private Redis Cloud database. Any AI agent — Claude Code, Cursor, your own scripts — can then ask architectural questions about your codebase via our API and get grounded answers in seconds, not minutes.

Languages we support

Python SUPPORTED
Full AST extraction. Functions, classes, async functions, methods. Ancestry tracking via static analysis.
TypeScript / JavaScript SUPPORTED
Full AST via TypeScript compiler. Functions, classes, components, interfaces, types. React components recognized.
C / C++ ON REQUEST
tree-sitter integration planned. Built when first hardware-domain pilot signs.
PowerShell ON REQUEST
tree-sitter integration planned. Built alongside C/C++ for systems pilot customers.

What it costs

Mode Best for Per query
Quick Lookups, simple questions, fast iteration ~$0.005
Standard Architectural questions, design discussions ~$0.10
Deep Security reviews, cross-cutting analyses ~$0.03 - $0.30

Ingestion costs are billed separately based on tokens used. Typical commit ingestion runs $0.01-0.05 per commit depending on number of changed symbols. Redis Cloud costs are paid directly to Redis (free tier available, paid plans start at ~$5/mo).

Common questions

Do you have access to my code?
Symbol code bodies and skill descriptions are stored in your Redis Cloud instance, which is provisioned in your org's name. We process code during ingestion and Q&A but don't retain it beyond your Redis database. Your code never trains anyone's model.
What's a "skill"?
A skill is a structured description of what a piece of code achieves — its role, what it interacts with, what's needed to use or extend it. Each function, class, and significant mechanism in your codebase becomes one. Skills are vector-indexed so retrieval finds intent, not strings.
How is this different from grep or semantic search?
Grep finds strings. Semantic search over raw code finds passages that look similar. DevLoops indexes what code does, not what it contains. Asking "where do we handle authentication?" returns the actual auth flow, not every file with the word "auth" in it.
What about repos that aren't on GitHub?
DevLoops works on any local git repo. The CLI runs on your machine and only ships extracted symbol metadata to our backend — not your remote URL or auth credentials.
Is this stable enough for production?
It's beta. The product is functional and we use it on our own codebase daily. We're onboarding early customers selectively and prioritizing accuracy over feature breadth. Expect rough edges in the UI and occasional ingestion retries.
Sign up → View pitch Home