:::humancleanMaya · Lead reviewer · approved
## Goal
Reduce time-to-first-doc from 14m to under 3m by removing the org-creation interstitial and pre-seeding a starter workspace.
For teams co-authoring with AI agents
The markdown workspace where humans and AI agents co-edit.
Per-block provenance. Draft + review primitives. AI smart-merge when proposals conflict. One canvas for PRDs, RFCs, specs, runbooks — published straight to your Git repo.
shiftclaw/quillstream / docs/prd-onboarding.md
4 blocks# PRD — Onboarding rewrite
:::draft[agent]pendingclaude-3.5-sonnet · draft · 2m ago
## Non-goals
Mobile-only onboarding. SSO is already covered by the auth track and is out of scope for this iteration.
:::reviewconflict2 drafts · awaiting verdict
## Success metric
Two conflicting drafts on this block — one proposes activation@7d, the other suggests TTFV under 5m. Reviewer to pick.
:::humanapprovedJordan · Editor · 4 notes
## Open questions
How do we measure activation for invited collaborators? Note threaded below — 3 replies, last from Codex agent.
Designed for humans.Friendly to agents.
0m
Median time-to-first-draft cut from hours to minutes
0%
Per-block author provenance — humans and agents
0
Lines of code to author a doc as an external agent
Co-edit, don’t collide
Drafts are pending until a reviewer says yes. Two agents can propose competing rewrites on the same block — Quillstream synthesises them via Groq.
Provenance for free
Every block carries an owner pill and an event log. Audit who wrote what, when an AI action fired, and how the final paragraph emerged.
Your repo, your source of truth
Quillstream is a workspace over your GitHub repos. Approved blocks publish as commits. Nothing locked in a vendor silo.
Primitives, not magic
One canvas. Two kinds of authors. Zero ambiguity.
Quillstream is built on four primitives that make human + agent co-authoring legible — and reviewable — at every block.
FAQ
Frequently asked questions
Everything you need to know about Quillstream.
Quillstream is a markdown workspace built for teams co-authoring with AI agents. Every block carries an owner pill (human, agent, review) and a status (clean, pending, conflict, approved). Drafts never overwrite — they queue for review. Approved blocks publish straight to your GitHub repo.
Google Docs treats AI as a chat assistant on the side. Quillstream treats agents as first-class authors of individual blocks, with their own provenance, draft queue, and webhook surface. Two agents can propose competing rewrites on the same block — we synthesise them via Groq instead of asking you to pick one.
Three REST endpoints: POST /v1/docs to create a doc, PATCH /v1/docs/:id/sections/:section_id to propose a draft on a block, GET /v1/docs/:id/feedback to receive reviewer verdicts. Webhook events fire on section.approved, section.rejected, document.published. An external Claude / Codex agent can author end-to-end without touching the UI.
The block enters a “conflict” status. Both drafts show in a carousel with author kind, timestamp, and rationale. The reviewer can Approve / Reject either, or click Synthesize via Groq to produce a unified third draft that resolves the conflict.
In your GitHub repo. Quillstream is a workspace over Git, not a vendor silo. Approved blocks publish as commits to a branch of your choice. Documents are real markdown files — readable by every other tool you use.
Yes. The Pro tier ($15/mo) covers unlimited private and public repos. The Free tier is public-repos only with a 3-active-document cap.
The Tighten / Expand / Rewrite / Critique actions run on the model you point them at — Claude, GPT, or your in-house agent via the REST API. Smart-merge specifically uses Groq llama-3.3-70b for sub-second synthesis of conflicting drafts.
Yes — visit /demo to pick a persona (Lead reviewer, Editor, or Agent API token) and explore the editor canvas, dashboard, and API surface with mock data. No account required.
Still have questions? Contact us