01
Sign up & create a doc
Create a free account and open your first document. The editor lives in the browser — live collaboration from the first keystroke, nothing to install.
Sign up freeLaunch notes — v2 editor
The new editor ships with collaboration that we think is pretty solidholds up under real concurrency. Every participant — human or agent — submits steps to one authoritative log.
Agents don't get a side door. They connect over the same live protocol as the browser, and their edits land as tracked suggestions that are applied immediatelyawaiting human review.
Rollout begins next week, starting with design partners.
Get started
No SDK, no glue code — the editor, the CLI, and the MCP server are already wired together.
01
Create a free account and open your first document. The editor lives in the browser — live collaboration from the first keystroke, nothing to install.
Sign up free02
One command plugs Claude Code, Cursor, or any MCP client into your workspace. Grab an API key from the command palette and you're wired up.
03
Ask the way you'd ask a colleague. The agent reads your workspace skills and drafts straight into the live document.
you → claude
“Draft a design doc for the billing rework — context, options, decision.”
04
The draft lands as suggestions, attributed to its author. Accept, reject, or roll back — that's the loop.
Decision: ship usage-based billing behind a flag.
How it works
Humans in the browser, agents over CLI and MCP, programs over the API — every interface is first-class, every edit is accounted for, and nothing lands without passing the review gate.
Humans
UIthe live browser editor
Agents
CLIMCPClaude Code, Cursor, pipelines
Programs
APItyped HTTP · OpenAPI spec
One document, edited by whoever got there lasteveryone at once — every edit a tracked, reviewable step in one authoritative history.
same live connection · same suggestion rules · same review gate
Exported
Shared
Published
01
AI agents read, draft, and revise over the same live connection as your browser — via CLI or MCP. First-class authors, not a bolted-on assistant.
02
Every agent edit lands as a tracked proposal, attributed to its author. Accept, reject, or roll back — nothing ships without your sign-off.
03
All authors converge on one authoritative history. Every revision is recorded, every conflict resolves, every word is recoverable.
Security
Letting agents write into your documents takes trust. We earn it twice: by running like a security-first company, and by keeping your documents in storage you control.
An editor your agents write through sits inside your supply chain — we build and run it accordingly, to the standards you'd audit us against.
Bring your own storage: point terser at a bucket you own and every document, snapshot, and step of history persists in your infrastructure. Leave anytime — your documents were never anywhere else.
Stored in someone else's cloudyour bucket.
Your bucket
s3://acme-docs/terser
your region · your keys · your retention
Deploy terser inside your own perimeter — nothing leaves your network. And it never falls behind: we keep your on-prem instance current with every release, so the latest features and security fixes land for you without drifting onto a stale fork.
your network
terser, self-hosted
running · v214
your hardware · your perimeter · latest release
The review gate
Agent edits arrive as tracked changes in the document — green for proposed text, red for proposed cuts — each attributed to its author and waiting on your verdict.
claude · agent
holds up under real concurrency
claude · agent
we think is pretty solid
claude · agent
awaiting human review
The agent surface
A CLI for scripts and pipelines, a hosted MCP server for Claude, Cursor, and friends — both thin wrappers over the same command core the UI uses. No side door, no stale copies.
$ terser read launch-notes
# Launch notes — v2 editor …
$ terser edit launch-notes --suggest
✔ 3 suggestions submitted · v214
$ claude mcp add terser https://terser.dev/api/mcp
✔ agents speak terser natively — CLI or MCP
The house style
Writing skills are guidance documents — a style guide, tone rules, a doc template — that agents consult before they draft. Author them once in the editor; every connected agent writes your way.
claude read terser://skill/style-guide before drafting
The sharing model
Invite a person by email, grant a whole organization in one row, or mint a link anyone can open — three grant kinds, one dialog, the same roles everywhere.
anyone with the link can view — no account needed
The history
Every step — human or agent, draft or verdict — appends to one authoritative, replayable history. The audit trail isn't a feature on top of the document; it is the document.
Use cases
Anywhere a document has to track a moving target. The agent keeps the draft moving; a human keeps it true.
Architecture calls get captured while the discussion is still warm — the agent drafts the decision record from the thread, options and trade-offs included. The log stays honest because writing it down costs nothing.
agent drafts the record from the thread → you sign the decision
adr-031-event-bus.md
claude · suggesting
Decision: adopt NATS over Kafka — ops cost dominates at our scale.
Wire an agent to your merges. It walks the docs, finds what each change made stale, and files the fix as suggestions — page by page, attributed and reviewable.
agent proposes updates on every merge → you approve what's true
docs/authentication.md
claude · suggesting
Sessions expire after 24 hours30 days, refreshed on use.
The agent reads the merged PRs and writes user-facing notes in your house style — terse, present tense, no commit hashes. Release week becomes a review pass.
agent writes from the merged PRs → you cut what doesn't matter
changelog.md
claude · suggesting
v2.4 — share a doc with your whole organization in one row.
Review feedback lands as tracked suggestions, the agent folds it into the draft, and the open questions stay visibly open until a human settles them.
agent folds in the feedback → you settle the open questions
rfc-usage-billing.md
claude · suggesting
Pricing: both models at launchmetered only — seats return in phase two.
While you're still firefighting, the agent reconstructs the timeline from logs and channels into a draft. The analysis and the action items stay human.
agent assembles the timeline → you own the lessons
postmortem-2026-06-04.md
claude · suggesting
14:02 deploy 8841 ships · 14:09 p99 ×4, pager fires
Agents that follow runbooks notice stale runbooks first. When a step doesn't match reality, the fix arrives as a suggestion instead of a Slack complaint.
agent flags drift as it works → you accept the patch
runbook-key-rotation.md
claude · suggesting
3. Run scripts/rotate.shterser keys rotate --all.
Word of mouth
We pointed an agent at the changelog and stopped dreading release week. The notes draft themselves overnight — I just show up with the red pen.
In Notion the AI rewrites the page and you hope you'd notice. Here every agent edit is a suggestion with an author on it — accept, reject, roll back. Legal signed off in a week.
Our decision log went from tribal knowledge to an actual log. The agent drafts the record while the discussion is still warm; we just sign it.
01 / 02
Pricing
The full editor, the CLI, and the MCP server are in every plan — you pay for the people, never the agents.
$0 free forever
The whole single-player product — your documents, your agents, and the full review loop.
most teams start here
$5 / seat / month
Share the workspace. Humans take seats — agents ride along free.
Custom
For organizations with their own rules about identity, storage, and audit.