Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

90 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

pairputer

Experimental — NOT for production. pairputer on AWS is for hands-on testing and exploration. Deploy only into an AWS account YOU OWN and control. Never into shared, customer, or production environments. You are responsible for the resources it creates and the cost they incur.

Stream a live Linux MicroVM into your AI chat. Video, audio, keyboard, mouse. Runs entirely in your own AWS account.

pairputer is a deployable substrate: it runs an interactive capsule (a Lambda MicroVM workload) and streams it inline into an AI chat client.

You get a live viewport, the model gets a controlled tool surface, and you're both driving the same machine. It suspends on idle (Freeze) and resumes on demand (Thaw), so you only pay while you're actually using it.

Connect two connectors, ChatGPT (web) and Claude (web), and pairputer works across web, desktop, and mobile for each, including Codex. Once connected, open the reference capsule with a single prompt:

Open workbench using pairputer MCP

Connector setup: docs/chatgpt.md · docs/claude.md.

The bundled reference capsule is the Pairputer Workbench: a disposable, resumable Linux dev desktop (browser, VS Code, terminal) that you and whichever frontier AI you prefer operate together.

A persistent/ folder holding your git identity, editor settings, and projects survives Freeze, Thaw, and even Trash; everything else dies with the VM. You and the AI can drop files into it from the desktop or over chat.

Why pairputer

  • Runs in your AWS account. No third-party SaaS holds your session, and no static credentials leave your machine.
  • True 1-click. Signed, digest-pinned public images plus a public capsule build context. Nothing to build locally.
  • Secure by construction. OAuth (Cognito PKCE), a private VPC data plane behind CloudFront + WAF, and cosign-signed images with SLSA provenance you can verify yourself.
  • Bring your own workload. The Workbench is the default. The substrate is capsule-agnostic, and any capsule deploys as a cartridge with substrate/deploy-capsule.sh <capsule-dir>.

Deploy it

1-click deploy instructions, region notes, and cost estimates

One click, in us-east-1. No tools, no clone, no Docker - everything builds in your account from signed public images.

Launch Stack

Click the button, enter one input: your email address (it becomes the super-admin account and receives the invite), and deploy. Everything else defaults to pairputer's signed public images and the Pairputer Workbench capsule. Behind the scenes it stands up Cognito, the MCP control plane (Bedrock AgentCore), a private CloudFront-fronted data plane, and builds the Workbench MicroVM image in your account. To customize the deploy (private images, your own VPC, durable storage, region), every parameter is documented in docs/1-click-advanced.md.

After it finishes, you get an admin invite email with your temporary password. Then add the pairputer connector in ChatGPT (web) and Claude (web) using your stack's McpEndpoint output (see docs/chatgpt.md / docs/claude.md) and play.

us-east-1 is the tested and recommended region. The template isn't hard-locked to it, but other regions are unverified and you're on your own there: the CloudFront-scope WAF only exists in us-east-1 (deploy elsewhere and it's skipped unless you pass your own WebAclArn), and Bedrock AgentCore + Lambda MicroVM availability varies by region.

Want to verify the images first? Run scripts/verify-images.sh, an offline cosign signature + SLSA check.

💸 What does it create, and what does it cost? See docs/1-click-cost.md: a complete inventory of every AWS resource and IAM role the 1-click deploys (each linked to its CloudFormation source) plus honest daily/weekly/monthly cost estimates. TL;DR: roughly $55-60/month always-on, about $0.60 per active hour of Workbench use, and near $0 while Frozen.

Options worth knowing

Deploy parameters worth changing from the defaults

Leave every parameter at its default for the standard deploy. Two are worth knowing about:

  • Bundle reference capsule (default on): ships the Pairputer Workbench so the substrate is useful out of the box. Turn it off for a bare substrate with no capsule.
  • Image source (default Public): the signed public images. Advanced users can set Private to run their own private-ECR images (leave the URIs blank to auto-copy ours in, verified first).

Remove everything

Teardown commands and what gets deleted
substrate/remove-cf.sh            # delete the stack (all nested stacks go with it)
substrate/remove-cf.sh --all      # also remove the artifact bucket + ECR repos
substrate/remove-cf.sh --all --yes  # no confirmation prompt

Cartridge capsule stacks (pairputer-capsule-*) are deleted first automatically. Each one's reaper terminates leftover MicroVMs and deletes its image, then the root stack tears down every nested stack in dependency order. Nothing is left running, so the bill stops.

Learn more

About

A disposable computer you and your AI share, streamed into your chat.

Topics

Resources

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages