Thesis Desktop

A universal lab for autonomous experiment driven machine learning.

Thesis LabFileEditViewRun

9:42 PM

δThesis

/gold_forecasting~/Desktop/gold_forecasting

New Chat

Regime baseline

Volatility probe

Gold price EDA

Launch experiment

Gold price EDA

Running-Supervised-thesis/exp/experiment-a51f376-a51f3769

Overview

Agents

Code

Metrics

Logs

Turn the long-run gold price file into a research note: regimes, return tails, volatility, and what model family you would try next.

Starting with the time axis, then segmenting macro regimes and stress-testing the return distribution before recommending forecast baselines.

Ran: gold EDA notebookexit 0

$python3 -c "import pandas as pd; df = pd.read_csv('monthly.csv'); print('Shape:', df.shape); print(df.describe())"

Gold price history (1833-2026) - Free Float era dominates

Gold forecast lineage

Greedy tree - 15 experiments - val mae

Best path

SOFTWARE SUITE

Run Thesis everywhere

Thesis LabFileEditViewRun

9:42 PM

Thesis Lab

New Chat

Docking sweep

ADMET screen

Scaffold split

Kinase hit discovery

Launch experiment

Kinase hit discovery

Running-Supervised-thesis/exp/kinase-docking-bio72c4

Overview

Agents

Code

Metrics

Logs

We need an ABL1 wet-lab shortlist by Friday. Which ligands are potent, selective, and least likely to fail ADMET?

I am merging docking, QSAR, and ADMET screens into one ranker, then checking scaffold diversity for the validation batch.

Ran: ABL1 ligand triageexit 0

$python workflows/screen_ligands.py --target ABL1 --assays ic50,admet --dock vina --top-k 128

Docking pose panel - hinge binder candidates cluster in the ATP pocket

Kinase discovery lineage

15 experiments - docking + ADMET rank

Lead path

Lab Interface

Thesis Desktop

An agent-first experience, with all the best parts of the editor for when you need them.

████████╗██╗  ██╗███████╗███████╗██╗███████╗
╚══██╔══╝██║  ██║██╔════╝██╔════╝██║██╔════╝
   ██║   ███████║█████╗  ███████╗██║███████╗
   ██║   ██╔══██║██╔══╝  ╚════██║██║╚════██║
   ██║   ██║  ██║███████╗███████║██║███████║
   ╚═╝   ╚═╝  ╚═╝╚══════╝╚══════╝╚═╝╚══════╝
Accelerating AI research
21:13:03 sh › Detected platform: macos-arm64
21:13:03 sh › Latest version: v0.22.10
21:13:05 sh › Downloading thesis binary...
21:13:06 sh › Installing binary to ~/.local/share/thesis/bin
21:13:06 sh ✓ Installation complete!
Run 'thesis --help' to get started.

Terminal

Thesis CLI

Run agents in any terminal or script, with integrations to your preferred IDEs.

curl -fsSL sh.thesislabs.ai/install-core.sh | sh

Thesis Cloud

https://compute.thesislabs.ai

New Chat

Grasp dataset audit

Slip recovery rollout

Dexterous picking policy

Launch experiment

Dexterous picking policy

Running-Supervised-thesis/exp/pick-policy-rob41a9

Overview

Agents

Code

Metrics

Logs

The picker still misses clear clamshells and soft bags. Find the failure modes and propose a policy update we can canary.

Replay mining comes first; then I will compare recovery-policy rollouts against the cycle-time guardrail.

Ran: warehouse replay evalexit 0

$python robotics/eval_policy.py --task bin_pick --objects transparent,deformable --rollouts 500 --metric success_rate

Grasp failure map - misses concentrate on glare and soft packaging

Picking policy lineage

15 experiments - success rate

Deployable path

Web

Thesis Cloud Compute

Kick off agents in the cloud from your browser, phone, or tools like Slack, GitHub, and Linear.

Open Thesis Cloud