For students & researchers
Free for .edu
Used by Researchers at
In Action
Memory
You open Sundial on a Tuesday. Sunny already knows you ran ablations Monday, that two failed (CUDA mismatch), and that Maya commented on the third.
Yesterday: ran ablations 1-5 on baseline_v2
ablation_2: failed (CUDA mismatch, see cell 14)
ablation_3: Maya flagged the loss curve (thread)
ablation_4: best F1 so far (0.87)
Todo: rerun ablation_2 on the smaller batch?
8:37
LaTeX
Ask in plain English. Sunny writes it, compiles it, fixes its own errors. Citations from Zotero. Equations rendered live. No Overleaf timeouts.
format this as a 3-column results table, group by model
8:37
\begin{tabular}{lcc} \toprule
Model & Acc. & F1 \\
\midrule
GPT-4o & 0.81 & 0.79 \\
Claude & 0.84 & 0.82 \\
Sundial & 0.86 & 0.87 \\
\bottomrule
\end{tabular}
chap2.tex
View diff
Done. Want me to sanity-check the F1 against
your eval logs?
8:38
Notebooks
Ask in plain English. Sunny writes it, compiles it, fixes its own errors. Citations from Zotero. Equations rendered live. No Overleaf timeouts.
In [12]: model.fit(X, y)
> ValueError: shape mismatch (32, 768) vs (32, 512)
8:37
Fix the loader, then share with Devon for review
8:38
lab_results.pyinb
View diff
Shared with Devon. "Fixed dim mismatch in cell 12. Re-ran cells 12–18. Loss looks healthier."
8:39
Features
Comparison
Before
5+
In another tab, no context
Local, dies with your laptop
15s+, times out
Paid add-on
Paywalled
Stuck in its own app
None
Sundial
1
In the doc, with context
Live in the doc, shared
Live preview
Built in, free
Free for .edu
Connects in one command
Persistent per workspace


