Your product knowledge is buried in a hundred different tools. Harvest it.
Sales notes in HubSpot, research in Notion, decisions in Linear, support tickets in Zendesk — every team holds part of the picture; nobody holds the whole. Harvest is the methodology. Winnow is the open-source tool that compiles it all into one living source of truth your team — and your AI builders — can actually use.
Harvest the field.
Winnow the grain.
Most product teams operate in speculation but treat speculation as fact. Harvest's contribution is making speculation visible, traceable, and resolvable. Three states. One source of truth. Every claim cited.
Drop in what you already write
Sales notes, research transcripts, decision memos, support tickets. Drag-and-drop a folder, point a webhook at it, or sync from Dropbox / Drive / OneDrive. Whatever you've got, in whatever format.
The wiki maintains itself
An LLM reads each piece of evidence, decides which pages should change, and proposes the edits. Reinforcements and contradictions with prior beliefs are surfaced — no silent rewrites. You review the diff before anything sticks.
Hand cited briefs to your editor
Recommendations are copyable, structured, and sourced. Hand them to Cursor, Claude Code, or your IDE — citations survive the paste, so the AI build stays grounded in the same evidence the team did.
Procurement adds three to four weeks to enterprise rollouts in regulated verticals.
If we expose the citation graph during onboarding, time-to-first- recommendation halves.
Free-tier users churn within seven days if they don't upload evidence in the first session.
The object of the verb is your team's own writing.
“Harvest” can pattern-match to data-harvesting — the surveillance-flavoured kind, where someone gathers up signals about your customers without their knowing. That isn't what this is. The object here is your team's own writing — sales notes, research transcripts, decision memos — first-party evidence you authored and already own. It's a farmer harvesting their own crops, not someone helping themselves to the field next door.
The full version — what enters the system, what doesn't, and what flows out — lives at where the data comes from, where it goes.
Engineering-grade. Quietly opinionated.
A reading-room product, not a dashboard product. Calm where competitors are loud. Editorial where competitors are utilitarian. Self-host or run in the cloud — same source of truth either way.
Everything is sourced. Nothing is assumed.
Each recommendation is a structured brief: a position, the decisions it implies, and every cited source. Click any citation to read the original. Demote it back to hypothesis if the evidence turns thin. Export the whole brief as a paste-ready prompt.
Built for product people who write things down.
- Senior product managers compiling evidence across squads.
- Heads of product running quarterly strategy reviews.
- Founders codifying what the team already believes — and what they don't.
- Engineering managers turning research into prioritised, sourced tickets.
Want to see what this looks like in your role specifically? How Harvest fits product teams →
Run Winnow on your own infrastructure.
A single command to install. Your wiki, your evidence, your model. The cloud version is identical to self-hosted — the only difference is who runs the database.
# install $ npm i -g winnow # initialise a workspace $ winnow init ./my-product ✓ evidence pool (0 records) ✓ wiki (5 foundation pages) ✓ recommendations (0 briefs) # compile a recommendation $ winnow compile "onboarding friction" validated [s/2025-q1/acme-call-7] hypothesis [h/2025-04/onboarding-friction] refuted [r/2025-03/free-tier-churn] → wrote recommendations/onboarding-friction.md
CLI, plugin, web — same source of truth.
Use Winnow from your terminal, from a Discord agent, from n8n, or from the web app. The wiki and citations are the contract; the surface is up to you.