import-ai/omnibox

What it does

OmniBox (小黑) is a cross-platform knowledge hub that combines capture, editing and retrieval. You save webpage content through a browser extension, upload PDFs, Word, PPT, MP3 and other files for parsing and indexing, or write in a Markdown editor that renders formulas, mind maps, flowcharts and similar diagrams. The indexed material then supports Q&A and writing grounded in both the web and your own database. Additional entry points include an iOS capture app, an iOS share action, and a WeChat bot that accepts files, links, videos, voice messages and chat text. It also ships users and teams, permissions, sharing, multi-tenancy and dark mode.

It suits individuals or teams who want one searchable home for scattered reading material and notes, and who are willing to run the service themselves.

Quick start

The README documents three routes. The hosted option is omnibox.pro, which supports email, Google and WeChat login. For a local instance the documented development flow is:

git clone --recurse-submodules https://github.com/import-ai/omnibox.git
cd omnibox
cp example.env .env
bash scripts/dev.sh up -d --build

The submodule flag matters because the frontend, backend, wizard and extension live in separate repositories. After the containers come up, open the local web address printed by the dev script. Deployment-specific steps live in the local deployment guide, and installing the capture extension is covered in the browser extension guide.

Constraints

• Local setup depends on Docker-style container tooling invoked by scripts/dev.sh; the exact prerequisites and exposed ports are only described in the linked deployment docs, not in the README.
• The bundled repositories (web, backend, wizard, browser extension) are versioned and released separately, so component versions may not move in lockstep.
• Python is the language of the main repository, but the README does not state minimum Python, database or model-provider requirements; check the deployment guide before provisioning.
• The license is Apache-2.0. No pricing, quota or data-retention terms for the hosted service are given in the README.

HotRepo's take

HotRepo has no daily ranking, star-curve sampling or Hacker News thread recorded for this repository, so there is no site-side popularity signal to read into. The observable facts are the project structure and scope: a split into four separately released components, a documented self-hosting path, and an unusually broad capture surface spanning browser, iOS and WeChat. That breadth is the thing to weigh when evaluating it — it means more moving parts to operate than a single-purpose note app, and the practical test is whether the deployment guide gives enough detail for your environment.