Polaris

Read the paper, while Polaris implements it.

The Runtime environment for any ML Paper.

Polaris Reads the paper, Plans the build, and Codes the implementation. Every paper gets a sandbox to run Benchmarks.

How it works
CLI now available

Install Polaris locally, bring your own model key, and run the TrueForge-backed paper reproduction pipeline from any terminal.

npm package ↗
Install
$ bun install -g polarisai
Run a PDF
$ polaris run --file paper.pdf --auto

No more being lost reading a paper, our ensemble of agents are your North Star.

READ -> RESEARCH -> PLAN -> CODE is our mantra 🧘‍♂️

01 / 04READ

Understanding the paper

  • Understanding Intent
  • Extracting Claims
  • Citation graph
Scroll to continue →READ
02 / 04RESEARCH

Trace the references

  • Citation retracing
  • Context reconstruction
  • Numbers cross-check
Scroll to continue →RESEARCH
03 / 04PLAN

Piece together all the contenxt

  • Module blueprint
  • Test scaffold
  • Risk review
  • User Approval and Suggestions
Scroll to continue →PLAN
04 / 04CODE

Run the proof

  • Sandbox spin-up
  • Dep install
  • pytest run
  • Reproduction verified
Done — keep scrolling for pricingCODE
Agent 1 of 4
polaris — read session

Pricing

Hobby

Limited
$1$5/ month
  • ✓READ → CODE pipeline
  • ✓Plan approve / reject
  • ✓Limited Feedback Loop
  • ✓1 Customisation

Pro

$20/ month
  • ✓Priority Execution
  • ✓0.5x GPU access (beta)
  • ✓Limited Train Jobs
  • ✓20 Feedback Loops
  • ✓Sandbox File Preview

MAX

$100/ month
  • ✓Highest priority GPU queue (beta)
  • ✓100 Train Jobs
  • ✓Full Sandbox Access
  • ✓Unlimited Customisations