Introducing Capsule — The Infra Framework for AI Apps
Capsule is a Python framework that provides infrastructure primitives for sandboxes, auth, session management, integrations, and payments.
Capsule is @supabase for AI apps. You get a powerful Python SDK to iterate fast, and one command to deploy to the cloud.
Using Capsule feels like using @beam_cloud or @modal. You don’t need to set up any infrastructure. There’s no Docker, just Python. The feedback loop is super fast. You deploy to the cloud with one command.
At Beam, we have a bird’s-eye view into the AI apps people are building. Over the past year, we've watched people build the same 5-7 infra primitives over and over again. Capsule provides all of those primitives using a single SDK.
@llom2600 originally built Capsule to prototype our own AI apps. We started selling some of these apps for real money and realized every app needs this, not just ours.
Give it a try! Curious to hear what people think. Do AI apps have their Supabase yet?
Day 2 of building a YouTube thumbnail maker app.
My friend and I found a newly released open-source image model, self-hosted it, and reduced costs 2x per thumbnail.
Let's fcking go!
Beam (@beam_cloud) is an open source serverless platform for AI apps.
Run GPU inference, background jobs, and sandboxes with ultrafast boot times and no vendor lock-in.
https://t.co/xlMRBl43Yr
Beam Sandboxes fully support Docker and Docker Compose 🚢
Sandboxes are containerized cloud environments for running user code, reinforcement learning experiments, and evaluating LLMs.
We launched Sandboxes earlier this year, and we kept getting the same question from developers:
Do you support running Docker inside the Sandboxes?
Even though we’ve always supported Docker images, running the Docker daemon inside sandboxes required modifying our container runtime. To ship this securely, we added gVisor support to Beam.
This unlocks a few things:
> Run Docker and Docker Compose in isolated cloud environments
> Multi-tenant security without exposing the host Docker daemon
> Fully reproducible environments (if it works locally, it works on Beam)
If you're building something that needs secure, isolated Docker environments, give this a shot!
Speed of iteration is the single biggest factor to developer productivity.
That’s why, 4 years ago, we decided to build Beam.
Earlier this year, we partnered with https://t.co/ygbF4647ir. They needed a way to serve a document extraction API to customers. And they chose Beam — not because it was cheap, not because it looked cool, not because it was some fancy new AI platform.
The founder, Adrian, is a technical guy.
He knows Kubernetes, he can set up an autoscaler, and he knows how to package ML models using Docker and FastAPI. But he didn’t want to spend time thinking about that stuff.
Since using Beam, they've closed several huge customers and 10x’d their document processing pipeline.
We’re thrilled to partner with them on their journey.
Introducing @vibekit_sh + Beam 🎉
VibeKit is the safety layer for running Claude Code, Gemini, and other coding agents.
You can now run Claude Code, Gemini, and other coding agents in secure cloud sandboxes with:
- Snapshotting and forking
- Custom Docker images
- Ultrafast boot times
- GPU support
- No timeouts
Get started --> `npm install @vibe-kit/beam`
Excited to announce the new Client class in the Beam SDK
This SDK update makes it easier to programmatically interact with your Beam apps.
You can:
- Retrieve tasks and deployments by ID
- Submit file inputs to your endpoints, without writing boilerplate code
- Retrieve task IDs immediately after running task_queue.put()
- Make your functions wait for task results
Check out the docs here: https://t.co/trt0EnsisQ
80% faster cold starts with one click
Just added a new checkpointing feature in your Beam dashboard:
- Go to any deployment
- Toggle "checkpoint enabled"
- Watch cold starts drop up to 80%
Works on models you've already deployed, no code changes needed.
That's it. Your users get ultrafast responses instead of waiting 30 seconds for your model to load into memory.
Live now in your dashboard.
Announcing 🔒 Spending Limits
Set a spending limit on Beam in 2 clicks:
1. Dashboard → Settings → Workspace Budget
2. Enter your monthly budget
We'll automatically pause your containers when you hit your limit.
It's live now and takes 30 seconds to set up.
Just launched Memory Snapshots for Sandboxes 📷
Memory snapshots let you freeze any Sandbox state:
- Run sandbox.snapshot_memory() on your running Sandbox
- Capture memory state as immutable artifact
- You can spin up identical environments instantly from that snapshot
It's perfect for testing code variations without losing your current state.
Or skipping 5 minutes of setup every time you create a new session.
Most RAG apps fail due to retrieval. Today, we'll build a RAG system that self-corrects inaccurate retrievals using:
- @firecrawl for scraping
- @milvusio as vectorDB
- @beam_cloud for deployment
- @Cometml Opik for observability
- @Llama_Index for orchestration
Let's go!
Prototyping on GPUs sucks.
You spend hours spinning up instances, waiting for images to build, deploying APIs, then tearing it all down.
We've used AWS, GCP, and Lambda Labs. None made it easy to quickly run code on a GPU.
Here's what finally fixed it.