Give your agents one real document to work on — not another copy in another chat. They draft, research and revise. You see exactly what changed, edit anything, keep every version, and publish the document you actually approved.
One document. Any number of agents. Every change accounted for.
An agent can produce a twenty-page report in minutes. Then the trouble starts. Ask for another pass and you get another copy. Correct a figure and it reappears two turns later. Edit the file yourself and the agent never sees your change. Hand it to a second agent and you've started another branch of reality. After enough iterations, nobody is sure which version is the version.
The document shouldn't live in the chat.
Chat is where work is instructed. Docgent is where the document lives. Every human and every agent works against the same canonical document — the one current state, not a fresh copy. Agents propose changes to that state. Humans review, edit and approve them. The approved state is where the next turn begins, for whoever or whatever takes it.
No competing attachments, chat outputs or "final-final-v3". There is always one current state.
See what an agent actually changed — sections, figures, recommendations, structure — without rereading the document.
Human edits and approvals become part of the document's history. The next agent starts from what was accepted, not from what it last generated.
An agent can rewrite more in thirty seconds than you can comfortably review in thirty minutes. Docgent turns each pass into a revision you can actually review: the difference between the current document and the proposed one — which sections changed, which figures moved, what was left alone — instead of another twenty pages to reread.
Review the change, not the whole document again.
Ask an agent to create, research or revise a document from wherever you already work.
It reads the current document and works against that exact state — not a copy.
Docgent shows the proposed revision and what changed. Edit directly, comment, accept, or send it back round the loop.
The accepted revision becomes the document. The next person or agent starts from there.
The twentieth revision should be as understandable as the first.
A real document isn't just paragraphs. It has key figures, recommendations, tables, evidence, citations, decisions and appendices — things with meaning.
The document you approve is the document that ships.
Where Track A is elected, Inkl will pay Publisher the Royalty Rate for each Premium Read in the relevant Quarter, subject to any Minimum Guarantee specified in Schedule 1.
Where a Minimum Guarantee is specified in Schedule 1, if the Premium Read royalty in any Quarter is less than the Minimum Guarantee, Inkl will pay the Minimum Guarantee for that Quarter.
Humans read, edit, comment and approve in a proper editor. Agents connect through the Docgent API or MCP to read the current state, pull relevant context and history, propose targeted or document-wide revisions, and respond to human feedback. Neither side has to impersonate the other — no agent clicking through a human editor, no human editing JSON.
Works with: Claude, ChatGPT, Cursor and custom agents via API.
Memoranda, valuations, due-diligence and deal materials that take dozens of substantive revisions before anyone signs.
Board papers, committee packs and strategy documents where accuracy, provenance and sign-off matter.
Deep research and recommendations produced quickly by agents, then shaped by accountable professionals.
Client deliverables where the person whose name is on the document must stand behind every line.
Consistent reports across clients, entities or periods without giving up editorial control or design quality.
Docgent is built on durable document primitives, not an opaque proprietary editing model. Your documents stay yours.
Documents remain readable, structured source files you can take with you.
Every revision has an identity. Every earlier state can be recovered.
Nothing becomes canonical without a human decision.
Agents work with documents programmatically instead of pretending to click through a human editor.
They already know how to write. Give them a document they can come back to — one that remembers what humans changed, what agents changed, what was approved, and what happens next.
Agents do the writing. Humans stay in control.