Harakumo hosts your app and gives it a database, file storage, sign-in, email, payments, domains and AI models under one account and one API key, and Claude can build and run all of it for you.
Needs Node 18 or later. Rather ask than type? Connect Claude at https://harakumo.com/mcp.
Each service below works today, and all of them sit behind the same account and API key. Where something is missing, its card says so.
Static sites on the edge and Node.js apps in containers, at your-project.harakumo.app. Deploy from GitHub, a folder, the CLI or a chat; a failed build never replaces what is live.
Limits: Server apps sleep after 15 idle minutes. Pushing to GitHub does not deploy on its own yet.
One JavaScript file on the edge with its own URL and no server to size. Write it in the dashboard or send it from the API.
Limits: 200 KB per function. No logs or scheduled runs yet.
Search for a domain, pay from your balance and it connects itself, or bring one you already own. DNS editing, auto-renew and transfers included.
Limits: Buy .com, .dev, .app, .io, .ai and .cloud today.
A code home inside Harakumo: numbered commits you can deploy, and roll back by deploying an older one.
Limits: Through the API and the AI connector only. No dashboard page or git push yet.
SQLite databases at the edge, queried with SQL over HTTPS or the SDK, with a console and .sql export in the dashboard.
Limits: 10 GB per database. Postgres, MySQL and Redis are not available yet.
Private file buckets. Browsers upload and download directly with signed links, so the bytes never pass through our servers.
Limits: Up to 5 GB per file. No public file URLs yet.
Vector collections for semantic search and RAG. Send text and we embed it, or bring your own vectors, then filter by metadata.
Limits: Up to 1,536 dimensions.
Memory for AI apps: key-value with expiry, chat history kept per user, and semantic recall by meaning.
Limits: Chat history keeps the last 1,000 messages per user.
10 models from Anthropic, OpenAI, Google, xAI, Moonshot and Meta behind one key, plus harakumo-1, which fails over between vendors. Streams, and speaks the OpenAI API.
Limits: Up to 8,192 output tokens per answer, rate limited per workspace.
A model, standing instructions and a chosen set of Harakumo tools, run on demand, with every step and its cost logged.
Limits: Tools are Harakumo’s own. A run finishes inside one request, up to 24 steps.
Add harakumo.com/mcp to Claude or any MCP client and run the platform by asking. It acts with your role, and every change lands in your audit log.
Limits: Tested with Claude; other MCP clients are not verified yet.
Email and password sign-in for your app’s users, with email verification, password reset and signed tokens you verify yourself.
Limits: No social login or multi-factor sign-in for your users yet.
Take card payments without opening a payment-processor account: create a checkout link, get paid, withdraw after a one-time identity check.
Limits: 3.9% + 30¢ per sale. US dollars, one-time payments. Earnings are held 7 days.
Transactional email in one API call, sent under your app’s name.
Limits: From your own domain or no-reply@harakumo.com, one recipient per email.
Video upload with streaming playback, and live channels: send RTMP in, play HLS out.
Limits: Video only, up to one hour per upload.
Not built yet, and on the list: Postgres, social sign-in for your users, email from your own domain, auto-deploy on git push, and function logs.
Harakumo speaks the Model Context Protocol. Add one address to Claude, or any MCP client, and run the platform by asking.
https://harakumo.com/mcpYou sign in, choose which workspace it may act in, and choose whether it may change things or only read. It can do what your role allows and nothing more, every change it makes lands in your audit log, and you can disconnect it at any time. Tested with Claude.
npm install @harakumo/sdk. The code below uses real calls; the output shows the shape of what comes back.
import Harakumo from '@harakumo/sdk';
const hk = new Harakumo({ apiKey: process.env.HARAKUMO_API_KEY });
const { project } = await hk.projects.create({ name: 'my-site' });
let { deployment: d } = await hk.deployments.uploadDir(project.id, './dist');
// Poll until the build settles.
while (!['ready', 'failed'].includes(d.status)) {
await new Promise(r => setTimeout(r, 3000));
({ deployment: d } = await hk.deployments.get(d.id));
}
console.log(d.status, d.preview_url);A minute and a half on what Harakumo does, and how an AI assistant drives it.
Call any model on the gateway with your Harakumo key, or ask for harakumo-1 and let it fail over between vendors. Existing OpenAI clients work when pointed at harakumo.com/api/ai/v1 with a Harakumo key.
Usage is charged at each model’s list price once your plan’s included amount is used.
Each number is the one the platform enforces. Anything we do not measure yet is labelled that way, not sold as a quota.
per workspace
For trying Harakumo and personal projects.
per workspace
For launching a real app with a small team.
per workspace
For many apps and larger teams, with the highest limits.
3.9% + 30¢ per successful charge, with no monthly fee. On a $29.00 sale you keep $27.57. Earnings are held 7 days before you can withdraw them.
Extra AI usage and domains are paid from a prepaid balance you top up, so there are no overage bills. Your card is charged only when you choose it.
Static sites and file storage grow with your traffic on the edge. Server apps and databases have real limits today. Here they are.
Served from the edge network and cached close to visitors, so a traffic spike needs nothing from you. Up to 500 files per site and 10 MB per file.
Node.js apps run in one small container each, built on a build machine with a 13-minute budget. An idle app sleeps after 15 minutes and the next visitor wakes it. There are no replicas or autoscaling yet, so put heavy traffic on a static front end with functions.
SQLite at the edge: 10 GB per database, 100 KB per statement, 100 bound parameters, 30 seconds per query and up to 100 statements per batch. Queries go over HTTPS, 600 requests per 5 minutes per workspace today. For more data, split it across databases.
Files move straight between the browser and storage through signed links (15 minutes by default), so uploads never queue behind our servers. Up to 5 GB per file. No public file URLs yet.
Hobby is free and needs no card. Upgrade a workspace when it needs more.
Start free