Where teams write docs with AI — and ship them

Write specs together.
Ship them with AI.

A multiplayer workspace for PRDs, reports, and proposals — write together, run AI workflows over your drafts, and ship a versioned, read-only link.

Source

product-brief.md
style-guide.mdPinned
interview-notes.md

Build

Structure the braindump
Critique vs. standards
Synthesize the clean draftRunning

Artifact

Payments PRDv3
forge.app/share/x7k2…
3 inputs · prompt v4 · run #42

Sound familiar?

Your team's most important documents are written in the worst possible way.

01

The copy-paste shuttle.

Doc in Notion, AI in a chat tab — paste in, paste out, lose the formatting, repeat.

02

One-player AI.

Five teammates, five private ChatGPT rituals. Nobody sees the prompt that worked.

03

Groundhog Day prompting.

Every new PRD starts from zero — same style guide, same prompts, re-pasted again.

04

Shipping means screenshots.

Stakeholders get a PDF export, and “which version did the client see?” has no answer.

The work is real. The tooling treats it like a throwaway conversation.

Is anyone building this?

Someone asked for exactly this.

In a widely-shared post, the CEO of one of the world's largest tech companies sketched the LLM workflow tool he wished existed — and asked: "Is anyone building this?"

"There is all this stuff, which I want to process/compute over in this iterated way, with some build artifacts being important/worth saving."
— a wishlist that made the rounds

That feeling is the product. Yes. We're building this.

  • 01

    "Manage a set of input files, plus general-purpose context"

    The Input Library — Markdown files with pin-to-context.

    Shipped
  • 02

    "Real-time collaboration"

    Live multiplayer editing — presence, cursors, comments, @mentions.

    Shipped
  • 03

    "Some concept of snapshots or VCS integration"

    Whole-workspace snapshots — freeze, diff, branch, restore.

    Coming soon
  • 04

    "Create/manage inference workflows and a stored set of prompts"

    A versioned Prompt Library plus multi-step Workflows.

    Shipped
  • 05

    "Access to general-purpose coding agents, not just chat models"

    Agent steps that run code — in an isolated sandbox.

    Shipped
  • 06

    "Compiled outputs / inference results, shareable externally"

    Versioned Artifacts with provenance and revocable share links.

    Shipped

The mental model

Separate what's durable from what's disposable.

Your documents, prompts, and process are your codebase. Forge treats them that way.

Source

version-tracked
launch-brief.mdv12
style-guide.mdPinned
critique.promptv4

Build

defined once
01Structure the braindump
02Critique vs. the pinned guide
03Synthesize the clean draft

run #42 · setup #1

Artifact

every run kept
Launch PRD
v1run #7
v2run #23
v3run #42share

next document — same workflow

01

Source

Documents and prompts, edited together and version-tracked — pin what should always be in scope.

02

Build

Workflows chain prompt and agent steps. Define the process once; run it on every new document.

03

Artifact

Versioned outputs with full provenance, shipped as stable read-only links.

Think make for knowledge work — wrapped in a friendly, multiplayer, document-shaped app.

How it works

From rough notes to a shipped document in four steps.

  1. Write together.

    Draft in the multiplayer editor — live cursors, comments, @mentions — with the always-relevant context pinned.

    Two cursors co-editing a spec, comment thread open
  2. Build your workflow.

    Chain prompt and agent steps; test any prompt on real content before wiring it in.

    The visual workflow builder with three chained steps
  3. Hit Run.

    Steps stream live and finish on the server — retry a failure from the failed step, not from scratch.

    The run console streaming a step, laptop lid optional
  4. Ship the document.

    Save a versioned artifact and send a stable read-only link — the recipient sees exactly that version.

    An artifact at v3 with its share link copied

What people build with Forge

The same process, every document — not the same setup, every time.

01

The product team

braindumpstructurecritiquePRD v3 ↗

Standards stay pinned; every feature runs the same pipeline. Stakeholders get a read-only link, not edit access.

02

The consultant

transcriptsummarizeactionsclient link ↗

New meeting? Drop in the transcript, hit Run, send a link — not an attachment. Every Friday.

03

The researcher / writer

notes (growing)synthesizedraft vN ↗

The synthesis re-runs as the pile of notes grows — every version kept, every output traceable.

Feature highlights

Built for work you'll run more than once.

structurecritiquesynthesizerunning

tab closed — run continues on the server

01

Durable runs

Server-side runs survive a closed tab; retry from the failed step.

style-guide.mdalways in scope
02

Pinned context

Always-in-scope inputs — stop re-pasting the style guide.

critique·v4( {{draft}} )

03

Prompt library

Versioned prompts with {{variables}} and a test panel.

$ agent run transform.ts

04

Agent steps

Steps that run code and transform files — in an isolated sandbox.

PRD v3 prompt v4 run #42 3 inputs

05

Provenance built in

Every artifact traces to its exact inputs, prompts, and run.

06

Real-time multiplayer

Live cursors, comments, and mentions — Figma-style.

Coming soon
M7

Workspace snapshots

Freeze, diff, branch, and roll back your entire workspace.

Why not just…

Why not just use Notion AI / ChatGPT / a script?

01

"…Notion AI?"

A writing assistant inside one doc. Forge runs shared, multi-step processes over your documents — and keeps the recipe as durable as the dish.

02

"…ChatGPT + copy-paste?"

Great for one-offs, invisible to your team. In Forge the context is pinned once, the prompts are shared — and the fiftieth run is one click.

03

"…a Python script?"

Reproducible, but invisible to non-developers. Forge is script-grade rigor with a multiplayer editor and one-click sharing.

Your work stays yours.

Workspaces isolated at two layers — app scoping plus database row-level security.

AI-generated code runs only in a locked-down external sandbox, never near your data.

Share links are unguessable, single-version, and revocable the moment you say so.

FAQ

Questions, answered.

Your next spec deserves better than a chat tab.

Write it with your team, run your process over it, ship a link — then do it again next week without rebuilding anything.

Get early access

No credit card. We'll email you when your workspace is ready.