npm install -g opencolab# Start the OpenColab onboarding flow
opencolab igniteAccelerating scientific discovery
Turn your research into an always-on autonomous lab.
OpenColab is building a system where humans set direction and research agents investigate, coordinate and run experiments on GPUs.
Basic architecture
A human and the professor agent set the direction together. The professor agent coordinates the workflow and delegates work to AI researcher agents. Each researcher can run with a different model and CLI, including Codex, Claude, Grok, Gemini, MiniMax, Kimi 2.5, and DeepSeek.
Quick Start
Research agent workflows
OpenAI challenge participation
Experiment 013: AR Latent Diffusion
AutoResearch, an OpenColab agent, ran over 60 experiments in a 1x H100 sweep for the Parameter Golf challenge, improving single-GPU validation BPB from the 1.2025 reference to a 1.1868 run within the 10-minute budget.
Read the experiment notesTelegram research request
Deep Research: JEPA literature review
From Telegram, the user asks an agent to research current advances in JEPA architecture. The agent summarizes over 35 papers, generates a LaTeX report with key architecture images and references, and returns the final PDF to the user.
What it does
Deep research swarm skills
Search, download, and summarize 100 papers in parallel, with grounded QA (Reasoning-based RAG), figure extraction, and block diagrams for scientific research workflows.
Multi-provider runtime support
Run researcher agents across OpenAI, Anthropic, Gemini, MiniMax, and xAI with the runtime that best fits the job.
Multi-project agent workspace
Coordinate multiple projects and agents from a local workspace with both CLI and Telegram control loops.
External GPU experiments
Launch experiment runs on external GPU servers (via Runpod) while keeping agent planning and execution organized locally.
LaTeX paper generation
Planned next: generate research papers in LaTeX format as part of the end-to-end lab workflow.