ResearchCraft Desktop
⬡ 25 credits
Research Topic
"Efficient diffusion for ImageNet top-5 accuracy vs ViT-B/16…"
Data File
imagenet_val.csv
Clarify →
Generate ⬡5
Q&A
✓ Topic refined
Metric: top-5 acc
Baseline: ViT-B/16
Dataset: ILSVRC
ℹ Sending generation request…
ℹ Code received: diffusion_exp.py
ℹ Data path injected
ℹ Attempt 1…
✗ ModuleNotFoundError: diffusers
ℹ Auto-fixing (1/3)…
ℹ Attempt 2…
✓ exit 0 · 6.1s
import torch, diffusers
from pathlib import Path
def run_experiment(data_path):
"/Users/me/imagenet_val.csv"
Same ActOn account · credits shared with the web app
Everything you need to
run real experiments.
The desktop app handles what a browser can't — local file access,
process execution, output streaming, and iterative auto-fix loops.
All AI work stays on the server so your credits always apply.
Your real data files
Browse your local CSV, JSON, or Parquet files. The app automatically
injects the real path into generated code — no manual edits required.
Live stdout / stderr streaming
Every print statement, warning, and traceback streams directly into
the execution log in real time — no polling, no refresh.
Up to 3 auto-fix retries
When code fails, the app sends the traceback to the server, gets a
patched version back, and retries automatically — up to 3 times.
Credits deducted as normal
Generation (5 credits) is the only paid step. Auto-fix retries and
paper search are free. Credits sync across web and desktop.
Secure token storage
Login tokens are saved to your OS keychain — you stay logged in
between sessions without exposing credentials in plain text.
Outputs saved to ~/ResearchCraft/
Every run is timestamped and saved locally — code, essay, and
paper list — so you always have a local copy of your work.