Edit Markdown, LaTeX or code files locally with all changes tracked in Sundial.
Supports local agents like Claude Code, Codex and Cursor.
One-click sharing to work with your team.
No card required. Free forever for .edu addresses.
This is the real app during a revision pass. You ask in chat, the agent edits the doc, and the change lands as a signed suggestion to keep or undo.
Claude Opus 5
Keep
Undo
✓ Kept · in history
Attention dominates the cost of long-context inference. We introduce a sparse decoding scheme that routes each query to a small, learned subset of keys, matching dense-attention quality on all four benchmarks we evaluate while decoding 3.1x faster at 32k tokens. The gain grows with context length and requires no retraining of the base model.
Per-token decoding cost grows linearly in context length, and attention quickly dominates as contexts reach tens of thousands of tokens. Prior work prunes heads or compresses the KV cache, trading quality for speed. We instead learn a routing function over the scores
which the decoder evaluates on a small routed subset of keys in place of the dense score matrix. Routing adds under 2% overhead and is trained jointly with a learned distance bias.
Efficient attention spans fixed sparse patterns, low-rank approximation, and cache eviction. Sliding-window and block-sparse layouts fix the pattern in advance, while eviction policies discard keys the model may later need. Learned routing instead adapts the pattern per query at decode time, keeping the full cache addressable.
Errors fix themselves while you type, completion knows your whole library, and a photo of a whiteboard becomes a table.
The agent watches as you type. When a link breaks, a number drifts out of sync with its table, or a typo slips in, it repairs the mistake right away, usually before you notice the problem. These fixes are applied live rather than proposed, but each is attributed like any other edit, so you can undo any of them.
It reads the whole project, not just the open file.
Start a citation and search your library without leaving the line. Sundial stays synced with Zotero and writes the reference entry itself, and the same completion knows every heading, figure, and term in your project.
A citation that matches nothing in your library gets flagged.
Drop in a whiteboard photo, a table screenshot, or a hand-drawn figure. It comes back rebuilt in the doc, a table as a table, a figure as a figure, checked and waiting as a suggestion.
Dictating a change works too; you review the diff.
Every AI edit is signed and reviewable, whether it was proposed as a suggestion or applied live. Keep or undo, one by one or in batches; history records who wrote what and when, human or model, down to the single edit.
Claude Opus 5 made 2 suggestions in this file
Our method achieves significant improvements over strong baselines across a wide variety of settings improves macro-F1 by 2.4 points on all four benchmarks at equal parameter count. The gain holds across all three seeds.
Claude Opus 5 · made 1 claim precise
It is important to note that much further work will be needed to fully explore We leave to future work scaling past 7B parameters and multilingual pretraining.
Claude Opus 5 · cut 1 filler phrase
Citations are read against the papers behind them, and your numbers stay consistent across files. Every finding arrives as a suggestion.
Each citation is read against the paper behind it, so when a claim drifts past what its source supports, you hear about it first. When a result changes, every file that repeats it (abstract, tables, conclusion) stays flagged until they agree.
Reviewer 2 runs these checks. Now you run them first.
Alpine glaciers have lost half their ice since 2000 (Zemp et al., 2019).
The suggestion flow sits on top of a serious editor. The day you need any of this, it is already there.
Documents render a live preview, Jupyter notebooks run inside them, and plain Markdown covers everything else. The agent executes, checks, and fixes its own errors in a sandbox.
Share a project with a private invite or a public link. Coauthors edit live, comment threads stay pinned to the text, and the free plan never caps seats.
Point your local agent at the workspace. It edits the same document through the same signed-suggestion review flow.
Agentfixing citationsFrontier or open source, pick per task. Review works the same no matter who writes.
Bring your work as-is. Two-way GitHub sync means your collaborators do not have to change anything.