Your machine, your keys
The agent harness is a local daemon. Your source never leaves your hardware, and you bring your own provider key: Anthropic, z.ai, Kimi or MiniMax, picked per difficulty level so easy cards don't burn the expensive model.
AI-assisted kanban · self-hosted agents
The AI does the analysis.
You do the building.
Ragondin Code is a kanban board where every card is picked up by an agent running on your machine. It reads the repo, asks the questions it actually needs answered, writes a phase-by-phase plan, and reviews your branch when you're done, including driving the live app in a browser to check that the thing works.
Still being built
Ragondin Code is in beta and under active development. It runs every day on our own products and a small fleet of ventures, which is how most of it got built, but it is not finished software. Expect rough edges, expect things to move, and expect to talk to a human when something breaks. If that sounds like the wrong time to adopt it, it probably is. If it sounds interesting, that is exactly who we want right now.
Access is by request while we are in beta. Email fabien@ragondin.dev and tell us what you are building and what you ship it with. You will get a reply from a person, not a form.
Created
Analyzing
Review
In progress
Code review
Done
The loop
From a one-line idea to a reviewed, merged branch
-
01
A card appears
You type one sentence and the AI writes the title. Or your app's users file it themselves, screenshot attached, through the intake SDK.
-
02
The daemon picks it up
It reads the linked repo, blocks to ask you the questions it genuinely can't answer, rates the difficulty 1–5, and produces a phase-by-phase plan.
-
03
You start it
Work lands on a
ragondin/<slug>branch. Drive it from your editor, or let a Claude Code session work the card through the agent API. -
04
It reviews the diff
Critical and high findings turn themselves into fix phases and bounce the card back, up to three cycles. A clean review is the only way out.
-
05
It uses the app
Give it a deployment URL and the usability agent brings the stack up, clicks through like a person, and reports what's broken. Strictly black-box: it never reads the source.
-
06
You ship it
Build & Submit compiles your Expo app into a signed build on your own Mac, with no cloud build service, then pushes it to TestFlight or the Play internal track.
What it does
Built for people who still write the code
Reviews that use the product
Code review reads the branch diff. Usability review launches the whole stack (ephemeral database, backend worktree, front end) and drives it with a browser, on iOS and Android simulators too.
A design doc that stays honest
Every org gets a living design document. When the shipped app drifts from the approved design, the reconcile pass flags the exact frame that moved: per-frame drift, not a vague "looks different".
Feedback straight from the app
Drop the feedback SDK into your Expo or web app and users file bugs with a screenshot attached. Submissions land in a triage feed, and nothing hits the board until an admin approves it.
It finds you when it's stuck
Blocking questions and terminal failures arrive as a Telegram DM with a Retry button. Answer from your phone and the job resumes. A morning digest tells you what the night shift got done.
One card at a time, on purpose
An org processes a single card until its merge request lands. No parallel branches off the same target, no merge nightmare. Extra cards queue and auto-start when the slot frees.
Pricing
At launchPay for the factory, not for the tokens.
Most AI builders resell you inference with a credit meter bolted on, and the work stops when the credits do. Ragondin Code sells the factory instead: the board, the agents, the reviews, the builds, the release. You plug in the model subscription you already pay for, and nothing runs out on a Tuesday afternoon.
These are the rates for when it ships. Ragondin Code is in beta today, so teams who come in now keep this pricing for a year after launch, and the first five pay nothing for three months.
You buy lanes. A lane is one org with one connected daemon: one card moving from analysis through review to release at a time. Everyone on the team can file cards and watch the board. Add lanes when you want more moving at once.
Lane
Start here
$99 / month
One org, one daemon, one card in flight. For a founder or a small team shipping a single product.
- Lanes
- 1
- People on the board
- Unlimited
- Model usage
- Your subscription
- Builds
- Your Mac, unmetered
Studio
$249 / month
Three lanes, for agencies and studios running several products at once. Every client gets its own org, its own daemon, its own queue.
- Lanes
- 3
- Extra lanes
- $79 / month each
- People on the board
- Unlimited
- Model usage
- Your subscription
Appliance
Custom / quoted
We provision the box, wire it to your repos, tune the loop to how your team actually works, and keep it patched. For teams who want it running, not set up.
- Lanes
- As many as you need
- Setup
- Done for you
- Build machine
- Provisioned + managed
- Support
- Direct, to a human
First five teams get three months free. In exchange: a call every couple of weeks about what is breaking, and permission to use your name once it is not.
Put your name downIn every plan
-
Plug in your own subscription
Anthropic, z.ai, Kimi or MiniMax, including coding-plan subscription keys, not just pay-per-token API accounts. Pick a model per difficulty level so a 1-pointer doesn't burn the expensive one. Keys are encrypted at rest and we never touch your usage bill.
-
Every step, not just the first one
Intake, analysis, the blocking questions, the difficulty rating, the phase-by-phase plan, the agent runs, and the queue that keeps exactly one card moving per org. The whole lifecycle, not a chat box bolted to a backlog.
-
Pull requests, opened and reviewed
A branch per card, the PR raised for you, the diff read locally and shown in-app file by file. Critical and high findings become fix phases and bounce the card back; conflicts come back as merge phases; a merge judge rules before anything lands on your target branch.
-
Reviews that actually use the app
The usability agent brings the whole stack up (ephemeral database, backend worktree, front end) and clicks through it like a person, on the web and on iOS and Android simulators. Black-box: it never reads your source.
-
Compiling your apps
Build & Submit turns a card into a signed
.ipaor.aabon your own Mac, using your repo's build script, with no cloud build service and no per-minute bill, then pushes it to TestFlight or the Play internal track. -
Managing the release
Every build tracked on one page: status, artifact, target, when. Store listing copy per locale with AI drafts and generated screenshots, saved to your store draft and submitted for review from the board. Signing and store credentials never leave your machine.
No credits, no build minutes, no per-review fee, no per-seat tax. The only meter anywhere in this is the model subscription you already own, and it bills you, not us. Self-hosted by design: the daemon is a process on your hardware, your source never leaves it, and the prices above are the whole of what we charge.
The pieces
Five parts, no magic
| Package | What it is | Built with |
|---|---|---|
| rc-express | API server | Express · MongoDB · Socket.IO |
| rc-react | Web app & PWA | Vite · React · Ant Design |
| rc-expo | iOS & Android app | Expo · React Native |
| rc-daemon | The local agent harness | Claude Agent SDK |
| rc-npm | Feedback & intake SDK | @ragondin/code-feedback |
Point it at a repo
Create an org, connect a repository, drop in a provider key, and file the first card. The board seeds a "Design the app" card so the AI has something to chew on while you read the docs.
It is a beta, so you will be early. That means the rough edges are yours to find, and the roadmap is still short enough that what you ask for can actually get built. Write to fabien@ragondin.dev and we will get you set up.