YOUR RESEARCH WORKSPACE

What are we investigating?

Start with a task, a puzzling result, or a hypothesis.
Work with your agent, then let a focused study run on its own.

A place to start

Choose an example and make it yours

1Connect an agentAPI endpoint or existing CLI

2Work through your taskChat, inspect files, prepare experiments

3Run a bounded studyEvidence, independent review, a clear result

YOUR COMPUTE

Machines

Connect your machines. Run experiments in parallel.

The research agent stays here. Connected machines run your experiments.

YOUR AGENT, YOUR MACHINE

API connections, if you need them.

Choose your agent and model inside each conversation. Installed CLI agents need no API setup here.

SIMJECTURE BENCH · EXPERIMENTAL

Compare scientific coding agents.

Prepare a versioned task, choose your existing agent in chat, and grade its numerical results independently.

Recorded solver outputs are included. These tasks measure finite diagnostic coding contracts; passing does not approve a scientific claim. Preparation makes no model request. Interactive chat is exploratory; use the Harbor export for enforced deadlines and controlled comparisons.

Your benchmark conversations

YOUR WORKBENCH

Research tools

Use what’s installed. Add a solver when your investigation needs it.

Start with Python. NumPy, SciPy, pandas and plotting are included. External solvers are optional.

Installed 1

On this computer · includes detected local builds

Available to add 0

Install directly or prepare with your agent

Installation and readiness checks tell you a tool can run. Each investigation still validates its scientific method. For tools on SSH workers, use .

＋ Attach files

Agent

Model

Reasoning effort

The agent can edit project files and run commands on this machine.

AUTONOMOUS RESEARCH

PREPARE IT TOGETHER

Let your agent shape the investigation.

Your conversation and files are the starting point. The agent fills the brief; you review it before launching.

Review or edit the study details

Agree on the investigation

EDITABLE BRIEF

Your agent fills these details from the conversation. You can change them here if needed. Each launched study keeps its agreed conditions.

Question or claim to testWhat evidence would answer it?Constraints and available inputs

When is the research complete?

Time budget (hours)

Research instruments Execution machines

Choose prepared workers for parallel experiments. Register a managed local worker in Machines to include it in a pool. Worker profiles define their own experiment environment.

Execution settings

Experiment environment

Cooperative execution is for a dedicated unprivileged account and does not isolate host networking.

Runs survive closing this page. They stop at the original deadline or an independently accepted outcome. You can pause or stop them earlier.

WorkspaceFiles & live runs

Simulations 0

Commands 0

Files

⌁

Your experiments, in view.

Ask your agent to run a simulation. Its live console and saved outputs appear here, beside the conversation.

Agent commands and their output appear here.

Project files 0

Open local folder

Copy large datasets here directly. Browser uploads support files up to 20 MB.

Research brief

When you’re ready, ask your agent to prepare an autonomous investigation.

Working principle

Explore freely. Keep the question explicit. Let recorded evidence and independent review decide the outcome.

What are we investigating?

A place to start

Start from a paper

Investigate a discrepancy

Find a first counterexample

Machines

Worker jobs

Add a machine