OpenBiliClaw is a local-first, open-source, self-evolving cross-platform AI content discovery agent that runs on your own machine, builds a private psychological profile of you over time, and then proactively goes out to find content for you across Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, Reddit, Linux.do, Bangumi, V2EX, Weibo, GitHub and the open web.

What it is

OpenBiliClaw is a Python 3.11+ project under the MIT licence that combines a local backend, a Chrome extension, and an optional client plugin for DeepSeek Harness. It lives in the personal AI agent space and treats the user's own machine as the centre of gravity: content discovery data is stored by default in a local SQLite database, and the agent keeps deepening its picture of you from your cross-platform usage, your feedback and your conversations. It can also be extended or rebuilt from source, and a Flutter mobile client in the separate OpenBiliClaw-mobile repository connects to the same local backend.

The concrete problem it solves is that platform recommendation feeds are opaque, remote and tuned for engagement rather than for one specific person. OpenBiliClaw replaces those platform-owned recommendation feeds with an agent that understands only you, runs only on your hardware, and can be corrected directly: marking something as liked or not interested, or saying so in chat, changes what it recommends next. Instead of being recommended to across many walled gardens one at a time, you get a single cross-platform discovery loop that you own and can tune.

Key capabilities

• Cross-platform discovery across Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, Reddit, Linux.do, Bangumi, V2EX, Weibo, GitHub and the open web from one agent.
• Local-first storage in SQLite by default, so the profile and history stay on your machine rather than in a vendor account.
• Tunable recommendation loop driven by likes, "not interested" signals and chat feedback that reshape later suggestions.
• Chrome extension installable from the Chrome Web Store under extension ID cdfjfkdjjhdaccbldipkjhpibnfbiamg, with automatic updates.
• Desktop backend installers for macOS (.dmg) and Windows (.exe) that stay resident in the menu bar or system tray, shipped as either a slim build that downloads the bge-m3 vector model on first launch or a -with-embedding build with that model (~1.1GB) bundled for offline use.
• DeepSeek Harness client plugin that adds a fourth column (recommendations, content library, chat, profile, settings) and registers 22 Agent Bridge tools so an agent inside DSH can read recommendations, answer probes and close the learning loop.
• Flutter mobile client in the OpenBiliClaw-mobile repository covering Android, iOS, web and desktop, with recommendations, chat, profile, favourites, watch later and 30-day history against the same local backend.

Who uses it and how

• Privacy-focused individuals who want a personal recommendation engine on their own laptop, with usage history and profile data never leaving the machine.
• Users in mainland China, supported through a Gitee mirror, an AtomGit source host, and a 123 cloud drive download page carrying the seven current packages including plugin and macOS/Windows builds with and without the bundled model.
• Developers who want to modify or deeply customise the stack, who paste a single instruction pointing at docs/agent-install.md into an AI coding assistant such as Claude Code, Codex CLI or Cursor to deploy the backend.
• DeepSeek Harness users who want personalised cross-platform content alongside their normal DSH work through the dsh-openbiliclaw plugin from the DSH plugin marketplace.
• Mobile-first users running the Flutter client against a backend hosted on their own desktop.

Getting started

Install the Chrome extension from the Chrome Web Store or from a zip in the latest release, then download the desktop backend for macOS (.dmg) or Windows (.exe) from the same release and choose either the slim build, which fetches bge-m3 on first run, or the -with-embedding build for offline use. Source-level deployment is done by giving an AI coding assistant the prompt that instructs it to follow docs/agent-install.md with Bash curl.

How it compares

The provided facts list no paid products that OpenBiliClaw replaces, and they name no directly comparable discovery tools. It therefore stands alone in this registry.

When to use it — and when not to

A self-hoster takes on a resident local backend, a local SQLite database and, on the slim build, a roughly 1.1GB vector model download, so a stable network or the offline -with-embedding package matters. Anyone who wants a zero-install hosted service, or who is unwilling to run a browser extension and a desktop process, should not pick this. Weaknesses visible in the facts: the Chrome Web Store build can lag a few days behind releases, coverage depends on the twelve platforms plus the open web that are listed, and the project carries 25 open issues with its most recent push dated 2026-09-17.