Interactive
Explore with your agent
Chat, run code and simulations, look at plots and decide the next step together. You stay in the loop.
Best for getting started, debugging, and working out what to ask.
YOUR RESEARCH LIBRARY
Knowledge
Find the work behind your next idea.
Connected sources
Folders refresh every minute while Simjecture is running. Conversations appear after a turn finishes.
Indexes text, code, PDF, PPTX, DOCX and notebooks. Hidden folders, dependencies, build caches and large binary outputs are skipped. Scanned PDFs need a text layer; figures are available in the original file.
One library for your agents
Connect an external agent to this workspace's enabled sources. These tools can search and read; they cannot change your files or start research.
MCP configuration
Search from a terminal
Run these commands on the machine that owns the folders. The browser server refreshes the index; when it is closed, refresh with:
PROJECT
Shared context
Available to conversations in this project. Each agent turn and study keeps its own snapshot.
This session is read-only. You can inspect projects and results.
What are we investigating?
Autonomous study
Let a study run on its own
Give it a claim and a time budget. It runs experiments and looks for counterexamples; an independent review checks the evidence before anything counts.
Best for “is this true?” questions you can state precisely.
Start in Interactive; when your claim is clear, switch to Autonomous study inside the conversation.
YOUR COMPUTE
Machines
Connect your machines. Run experiments in parallel.
The research agent stays here. Connected machines run your experiments.
Worker jobs
Detailed receipts
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 BENCHMARK RESULTS
Model leaderboard
Measured scientific coding tasks. Compare completion, time and API-equivalent cost.
Task leaderboard
Finish time ranking
Median verified finish time · shorter bars are better
Cost ranking
API-equivalent cost per attempt · shorter bars are better
Red marks unfinished runs. They have no verified finish-time rank; cost bars still show recorded spending, including labelled lower bounds. Configurations without inference are omitted.
Cost–time Pareto plot
Upper left is better. Dashed lines connect the same model's reasoning efforts.
Published API tariffs
Your runs & community contributions
Test any supported model with your installed agent or API connection. Local and imported runs appear in Community & local. Official results are maintained by Simjecture.
Prepare or grade a task
Your benchmark conversations
Latest numerical grade
EXPLORE YOUR WORKBENCH
Research tools
Find the right tool for your next investigation. Open a card to learn more.
Setup activity 0
Live installs, readiness checks and setups that need attention
On your workbench 0
Installed tools and detected local builds
Discover more 0
Open a tool to explore setup options
No matching tools
Try a different keyword or category.
Installation and readiness checks tell you a tool can run. Each investigation still validates its scientific method. For tools on SSH workers, use .
Continuing an earlier investigation ↗
Talk through the next step in chat; your agent then drafts its brief.
How an autonomous study works
It runs on its own after you approve a plan. Closing this page does not stop it.
- DiscussTalk the question through in Interactive, or let Grill me interview you.
- Review the briefYour agent writes the claim, the evidence needed and a time budget. Edit anything.
- LaunchExperiments run within the budget and look for counterexamples.
- Read the reportAn independent reviewer checks the evidence; your agent explains the result in the chat.
AUTONOMOUS STUDY
STEP 1 OF 2
Plan the study with your agent
Your agent uses this conversation and its files to write a study brief. Nothing runs until you review and start it.
STEP 2 OF 2 · REVIEW AND START
When the study finishes, your agent explains the report here in the conversation. Each study keeps its own results.