Skip to main content

2 posts tagged with "typescript"

View All Tags

Build and Publish Serverless Functions with GitHub Actions

Β· 7 min read


One of the principles that shaped Juno from day one was the idea of building apps with full ownership β€” no hidden infrastructure, no opaque servers.

No hypocrisy either.

If developers are encouraged to deploy code in containers they control, it feels inconsistent to rely on centralized infrastructure β€” like AWS or other Web2 cloud providers β€” to manage deployment pipelines or run the platform. With the exception of email notifications, Juno currently runs entirely on the Internet Computer β€” and that's a deliberate choice.

That doesn't mean being stubborn for the sake of it. It just means trying to push things forward without falling back on the old way unless absolutely necessary.

At the same time, developer experience matters β€” a lot. It shouldn't take a degree in DevOps to ship a backend function. Developers who would typically reach for a serverless option should be able to do so here too. And for those who prefer to stay local, it shouldn't feel like a downgrade β€” no one should be forced into CI automation if they don't want to.

That's why the new GitHub Actions for serverless functions are now generally available β€” for those who want automation, not obligation.


πŸš€ Automated Deployments, No Compromise​

With the latest release, it's now possible to:

  • Build serverless functions written in TypeScript or Rust
  • Automatically publish them to a Satellite
  • Optionally propose or directly apply upgrades

All within a GitHub Actions workflow. No manual builds, no extra setup β€” just code, commit, and push.

This makes it easier to fit Juno into an existing CI/CD pipeline or start a new one from scratch. The logic is bundled, metadata is embedded, and the container is ready to run.


πŸ” What About Security?​

You might ask yourself: "But what about the risk of giving CI full control over my infrastructure?"

That's where the improved access key (previously named "Controllers") roles come in.

Instead of handing over the master key, you give CI just enough access to do its job β€” and nothing more.

Here's how the roles break down in plain terms:

  • Administrator – Full control. Can deploy, upgrade, stop, or delete any module. Powerful, but risky for automation. Might be useful if you're spinning up test environments frequently.

  • Editor (Write) – Ideal for CI pipelines that deploy frontend assets or publish serverless functions. Can't upgrade those or stop and delete modules. A good default.

  • Submitter πŸ†• – The safest option. Can propose changes but not apply them. Someone still needs to review and approve in the Console or CLI. No surprises, no accidents.

Use Editor for most CI tasks β€” it gives you automation without opening the blast radius.

Prefer an extra layer of review? Go with Submitter and keep a human in the loop.


🧰 Local or CI: Your Choice​

Nothing changes in the approach for developers who prefer local development. The CLI remains a first-class tool for building and deploying.

All the new capabilities β€” from publishing functions to proposing or applying upgrades β€” are available not just in GitHub Actions or the Console UI, but also fully supported in the CLI.

In fact, the CLI has been improved with a neat addition: you can now append --mode development to interact with the emulator. This allows you to fully mimic production behavior while developing locally. And of course, you can also use any mode to target any environment.

juno functions upgrade --mode staging
juno deploy --mode development

πŸ›°οΈ Satellite's CDN​

While building serverless functions was never an issue, enabling GitHub Actions to publish and deploy without giving away full control introduced a challenge. How do you let CI push code without handing it the keys to everything?

That's where the idea of a sort of CDN came in.

Each Satellite now has a reserved collection called #_juno/releases. It's like a staging area where CI can submit new WASM containers or frontend assets. If the access key has enough privileges, the submission is deployed right away. If not, it's stored as a pending change β€” waiting for someone to approve it manually via the Console or CLI.

This builds on the change-based workflow that was added to the Console last year. Funny enough, it brought the Console so close to being a Satellite itself that it became… basically a meta example of what you can build with Juno.

And here's the cherry on top: because there's now a CDN tracking versions, developers can rollback or switch between different function versions more easily. A new CDN tab in the Console UI (under Functions) gives you access to all past versions and history.


πŸ–ΌοΈ Frontend Deployments, Too​

Frontend deployment now benefits from the same change-based workflow. By default, when you run juno deploy or trigger a GitHub Action, the assets are submitted as pending changes β€” and applied automatically (if the access key allows it).

Want to skip that workflow? You still can. The immediate deployment path remains available β€” handy if something fails, or if you just prefer to keep things simple.


πŸ› οΈ GitHub Actions for Serverless Functions​

Alright, enough chit-chat β€” here's how to publish your serverless functions on every push to main, straight from CI:

publish.yml
name: Publish Serverless Functions

on:
workflow_dispatch:
push:
branches: [main]

jobs:
publish:
runs-on: ubuntu-latest
steps:
- name: Check out the repo
uses: actions/checkout@v4

- uses: actions/setup-node@v4
with:
node-version: 22
registry-url: "https://registry.npmjs.org"

- name: Install Dependencies
run: npm ci

- name: Build
uses: junobuild/juno-action@full
with:
args: functions build

- name: Publish
uses: junobuild/juno-action@full
with:
args: functions publish
env:
JUNO_TOKEN: ${{ secrets.JUNO_TOKEN }}

🌸 One Action, Two Flavors​

Noticed the @full in the previous step?

That's because the GitHub Action now comes in two flavors:

  • junobuild/juno-action or junobuild/juno-action@slim – perfect for common use cases like deploying frontend assets or running simpler CLI tasks. No serverless build dependencies included, so it's faster and more "lightweight" (relatively, it still uses Docker underneath...).

  • junobuild/juno-action@full – includes everything you need to build and publish serverless functions, with Rust and TypeScript support. It's heavier, but it does the job end to end.

The right tool for the right job. Pick what fits.


🧭 Where This Is Going​

This release isn't just about smoother deployments β€” it's a step toward making Juno feel like real infrastructure. Though, what is β€œreal infrastructure” anyway? Whatever it is, this one doesn't come with the usual baggage.

Developers get to choose how they ship β€” locally or through CI. They get to decide what gets deployed and who can do it. They're not forced to rely on some big tech platform for their infra if they don't want to. And thanks to the new CDN and access control model, fast iteration and tight control can finally go hand in hand.

If you've been waiting for a way to ship backend logic without giving up on decentralization β€” or if you just like things working smoothly β€” this one's for you.

Go ahead. Build it. Push it. Submit it. Ship it.

To infinity and beyond,
David


Stay connected with Juno by following us on X/Twitter.

Reach out on Discord or OpenChat for any questions.

⭐️⭐️⭐️ stars are also much appreciated: visit the GitHub repo and show your support!

Announcing Serverless Functions in TypeScript

Β· 5 min read


One of the goals with Juno has always been to make building decentralized, secure apps feel like something you're already used to. No weird mental models. No boilerplate-heavy magic. Just code that does what you expect, without touching infrastructure.

And with this release, we're taking another step in that direction: You can now write serverless functions in TypeScript.

If you're a JavaScript developer, you can define backend behavior right inside your container. It runs in a secure, isolated environment with access to the same hooks and assertions you'd use in a typical Juno Satellite.

No need to manage infrastructure. No need to deploy a separate service. Just write a function, and Juno takes care of the rest.

Cherry on top: the structure mirrors the Rust implementation, so everything from lifecycle to data handling feels consistent. Switching between the two, or migrating later, is smooth and intuitive.


✨ Why TypeScript?​

Rust is still the best choice for performance-heavy apps. That's not changing.

But let's be real: sometimes you just want to ship something quickly. Maybe it's a prototype. Maybe it's a feature you want to test in production. Or maybe you just want to stay in the JavaScript world because it's what you know best.

Now you can.

You get most of the same tools, like:

  • Hooks that react to document or asset events (onSetDoc, onDeleteAsset, etc.)
  • Assertions to validate operations (assertSetDoc, etc.)
  • Utility functions to handle documents, storage, and even call other canisters on ICP

The JavaScript runtime is intentionally lightweight. While it doesn't include full Node.js support, we're adding polyfills gradually based on real-world needs. Things like console.log, TextEncoder, Blob, and even Math.random β€” already covered.


πŸ” Designed for Interop​

The approach to writing serverless functions in Rust and TypeScript is aligned by design. That means if you outgrow your TS functions, migrating to Rust won't feel like starting from scratch. The APIs, structure, and flow all carry over.

Start scrappy, scale gracefully.


πŸ‘€ A Taste of It​

Here's how simple it is to react to a document being created or updated:

import { defineHook, type OnSetDoc } from "@junobuild/functions";

export const onSetDoc = defineHook<OnSetDoc>({
collections: ["posts"],
run: async (context) => {
const data = context.data.data.after.data;
console.log("New post created:", data);
}
});

And here's for example how you'd validate a document before accepting it β€” using Zod for a clean, readable schema:

import { z } from "zod";
import { defineAssert, type AssertSetDoc } from "@junobuild/functions";
import { decodeDocData } from "@junobuild/functions/sdk";

const postSchema = z.object({
title: z.string().min(5, "Title must be at least 5 characters long")
});

export const assertSetDoc = defineAssert<AssertSetDoc>({
collections: ["posts"],
assert: (context) => {
const data = decodeDocData(context.data.data.proposed.data);
postSchema.parse(data);
}
});

That's it. No APIs to expose, no infra to manage. Just code that runs where your data lives.


πŸ–₯️ Local-First Development​

Alongside TypeScript support, we've rethought the local development experience.

Instead of providing a partial local environment, the mindset shifted to mimicking production as closely as possible.

You still get a self-contained image with your Satellite, but now you also get the full Console UI included. That means you can manage and test your project locally just like you would on mainnet.

See this other blog post for all the details.


🧰 Zero Tooling, Just Code​

Here's the beautiful part: even though your serverless functions are written in TypeScript, they're bundled and embedded into a Satellite module that's still compiled in Rust behind the scenes.

But you don't need to install Rust. Or Cargo. Or ic-wasm. Or anything that feels complicated or overly specific.

All you need is Node.js and Docker. The container takes care of the rest: building, bundling, embedding metadata and gives you a ready-to-run Satellite that runs locally and is ready to deploy to production.

In short: just code your functions. The container does the heavy lifting.


πŸ“‘ Already in the Wild​

This isn’t just a feature announcement β€” serverless functions in TypeScript are already live and powering real functionality.

I used them to build the ICP-to-cycles swap on cycles.watch, including all the backend logic and assertions. The whole process was documented over a few livestreams, from setup to deployment.

If you're curious, the code is on GitHub, and there’s a playlist on YouTube if you want to follow along and see how it all came together.

A screenshot of cycles.watch that integrates with OISY and use serverless functions in TypeScript to swap ICP for cycles


▢️ Try It Out​

We've put together docs and guides to help you get started. If you're already using the Juno CLI, you're just one juno dev eject away from writing your first function or start fresh with npm create juno@latest.

To infinite and beyond,
David


Stay connected with Juno by following us on X/Twitter.

Reach out on Discord or OpenChat for any questions.

⭐️⭐️⭐️ stars are also much appreciated: visit the GitHub repo and show your support!