PMOS
PMOS — Product Management OS

The PM workspace that reads the web for you.

PMOS pulls signals from the web, clusters them into feature opportunities, writes PRDs, and ships them to your tracker. The layer you used to fill with tabs and copy-paste.

The pipeline

Signals → Evidence → Suggestions → PRDs → Issues.

Each stage is automated with LLM agents and editable by the PM. You stay in control — PMOS does the undifferentiated work: reading, summarizing, tagging, synthesizing.

01

Signals

Web · social · GitHub

30+ collectors pull from Reddit, HN, arXiv, Product Hunt, company blogs, app stores, GitHub releases, YouTube, and more.

02

Evidence

Labeled items

A pre-filter pass tags each item with labels (pain point, feature request, trend, bug report) and writes a short summary.

03

Suggestions

Clustered features

Aggregate synthesizes labeled items into feature opportunities. Each suggestion links back to the signals that back it.

04

PRDs

Specs

Auto-generated markdown PRDs with competitor research baked in. Edit, preview GFM, then send.

05

Issues

Delivery

One click creates a GitHub issue on the configured repo; URL persists back on the suggestion.

Modules

Seven screens that replace a workday of tabs.

Every module is scoped to a product. Switching products in the sidebar re-scopes everything — sources, reports, suggestions, PRDs.

Discover

What's new in the world for your product's domain — research + industry + trends sources.

Listen

What users, competitors, and the community are saying — social, forums, reviews.

Compare

Competitor websites, repos, app stores, release notes. Report contrasts their moves vs yours.

Sources

Per-product source catalog with inspect, collect-now, pre-filter, and auto-discovery helpers.

Aggregate

Cross-module synthesis — clustered suggestions with evidence cards and batch history.

PRDs

Auto-save markdown editor, Edit/Preview tabs, and one-click GitHub issue creation.

Under the hood

Built for real PM work, not demos.

Claude Agent SDK under the hood. Local-first SQLite. 30+ collectors. SSE streaming for every long-running job. No per-token cost beyond your Claude plan.

App Router, route groups, streaming, SSE.Runs through the local claude CLI using your subscription — no API key required.All data local at backend/data/discover.db. Idempotent migrations on boot.~70 routes, SSE for every long-running job, clean Python module per collector.Reddit, HN, arXiv, Product Hunt, HuggingFace, Google Trends, app stores, GitHub, YouTube, X, LinkedIn, RSS — all pluggable.Cheap Haiku pass labels + summarizes every collected item before it reaches a report.Fine-grained PAT per product; one click creates an issue, duplicate sends blocked.Every generate, analyze, and aggregate emits granular progress events.Dark-first UI, tuned light mode, motion-reduced aware.Dockerfile, fly.toml, Procfile — or run make dev locally.Agents get real Google-style search without an external key.Token stored masked, shown as •••• + last 4. github_token_set boolean exposed in API.
Built on

Built on

Next.js 16FastAPISQLiteClaude Agent SDKTailwind 4shadcn/uiGSAPLenis