Sundial is a collaborative editor where AI edits arrive as signed, reviewable suggestions, and agents check your facts, citations, and consistency while you write. Checking is cheap here, so you can hand off more. Everything you ship is still yours.
Today we're shipping the biggest update since launch. Field Guide 2.0 rebuilds the planner around one idea: your plan should update itself when the world changes.
Rerouting is instant now. The new sync engine is a lot faster than the old one, and we think most people will notice the difference. The new sync engine rebuilds a full itinerary in under two seconds, down from forty. Fast enough to run on every edit.
Claude Opus 5 · suggested just now
Keep
Undo
✓ Kept · in history
Early-access teams have been living on 2.0 for six weeks. Below: what's new, what's moved, and Priya what we're retiring.
Every AI edit lands in your document as a signed, reviewable suggestion. Keep it or undo it, one by one or in batches. History remembers who wrote what, human or model, at the level of the edit itself.
Claude Opus 5 made 2 suggestions in this file
I run the community newsletter for a 400-unit co-op, which has given me a great deal of very valuable experience in many aspects of communication where a late issue means 400 people miss the water shut-off notice. Deadlines are not abstract to me.
Claude Opus 5 · rewrote 1 clause
Before that I spent two years leveraging synergies across cross-functional stakeholders translating engineering updates into notes residents actually read, which is the same job as this one, at a smaller scale.
Claude Opus 5 · removed 1 cliché
Agents read your claims against real sources and your citations against the papers behind them. When two files disagree, you hear about it before your readers do. Each finding arrives as a suggestion, and nothing changes without your say.
As you write, Sundial reads your claims and underlines the ones that do not hold up, backed by sources. Accept the correction or overrule it: the checker learns which sources you trust.
Runs on markdown and LaTeX alike.
The Amazon rainforest produces 20% of the world's oxygen and is often called the planet's lungs.
Sundial reads your citations against the papers behind them. When a claim drifts past what its source supports, you hear about it before your readers do.
Reviewer 2 runs this check. Now you run it first.
Alpine glaciers have lost half their ice since 2000 (Zemp et al., 2019).
One number changes and three files quietly disagree. Sundial reads across the workspace and flags claims that no longer match, before someone ships the wrong one.
Runs across every file in the workspace, not just the open one.
Every plan starts with a 14-day free trial, no card required.
Select a passage and fix it in place, or type a slash command where you need the result. Whatever comes back lands as a suggestion, in the same review flow as everything else.
Select a sentence that is not landing and click. Sundial offers four rewrites that keep your meaning, and you pick the one that sounds like you. Built with second-language writers in mind, used by everyone on a deadline.
In the recent months, the community garden knew a fast growth of participation. Saturday sessions are now weekly.
Before you publish, Sundial runs your draft against Pangram, the same AI detector your readers, editors, and professors reach for. If it reads like a machine wrote it, one click rewrites and re-verifies, pass after pass, until it does not.
Adversarial by design: the detector keeps you honest.
Describe the figure you need, or let Sundial read the surrounding section and propose candidates. Pick one and it lands in the doc, sized and captioned. No tab-switching to a prompt box.
Tidal locking is why the Moon always shows Earth the same face.
v1
v2
v3
v4Select a passage and drag. Sundial stretches or compresses it while keeping every claim intact: summary-length to talk-length in one motion, and back.
It is the same editor in every tab; only the files and the collaborators change.
Select a sentence that isn't landing and pick from four rewrites. Your choice, your phrasing on the page.
Nobody warns you that the first year is mostly failure. In the recent years, the interest for growing your own food knew a fast growth. The tomatoes died anyway.
Your notes stay markdown files you own. Capture a thought in chat and it lands in the right note, as a suggestion.
Fiction
Piranesi · finished in March
The Overstory · Sam's rec, start with the audiobook
On writing
Several Short Sentences About Writing · reread ch. 3
Quick capture
✕
Added it under Fiction in your reading list.
Claude Code and Codex join the document as collaborators. Their edits land as reviewable diffs, with history you can roll back.
Install with npm i -g sundial-cli, then run sun login once.
The sync daemon watches your workspace and pushes on every save batches pushes with a two-second debounce (--debounce to tune).
Conflict rules are documented in docs/sync.md.
Write the manuscript next to the notebook that produced its numbers, with an agent that can actually run both.
Live multiplayer for people and agents in the same document, with review that scales past one suggestion at a time.
Ship the beta to the waitlist on the 14th, then open signups a week later if the error budget holds.
Support rotation: we'll figure it out during the week Maya takes weekdays, Jae takes the weekend, and the agent drafts first replies for review.
Pricing copy is drafted in pricing.md. Claude Code refactored the tables from the spreadsheet export overnight.
The suggestion flow sits on top of a serious writing tool. Most days you will not think about any of this. The day you need it, it is already there.
The agent runs your notebooks, compiles your LaTeX, installs packages, and fixes its own errors in a sandbox of its own before anything touches your doc.
adding natbib, retrying…Point your local agent at a workspace and it becomes a collaborator: same document, same signed suggestions, same review flow as everyone else.
Agentfixing citationsFrontier or open source, pick per task. The review rules stay the same no matter who is writing.
LaTeX with live preview and Zotero citations. Jupyter notebooks that run inside the document instead of beside it. Markdown for everything else.
| Chatbot + paste | Autocomplete editors | Sundial | |
|---|---|---|---|
| Where AI text lands | In a chat window, pasted over your draft | Straight into your file | As a suggestion in the doc, signed by its author |
| Reviewing a big pass | Re-read everything, diff by eye | Re-read everything | Step through the diff: keep or undo each edit |
| Who checks the claims | You do, in another tab | No one | Agents check facts and citations as you write |
| Undoing one AI edit, a week later | Gone | Dig through snapshots | Every edit is in history; undo just that one |
| Checking its own work | You are the sandbox | No | Runs the notebook, compiles the LaTeX, cites sources |
The models will keep getting faster and cheaper. Whether your work gets better or just longer is decided by your editor.