OpenBiliClaw: Local-First Cross-Platform AI Content Discovery Agent
A local AI agent that deeply understands you and proactively discovers content across Bilibili, Xiaohongshu, Douyin, YouTube, X, Zhihu, and more.
- Source repo
- whiteguo233/OpenBiliClaw
- Stars
- ★ 3.5k
- Last updated
- 2d ago
- License
- MIT
- Primary language
- Python
- FA score
- 63/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformChatGPT · Codex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Content enthusiasts who want to escape platform algorithms and discover novel, cross-domain content aligned with deep interests across Bilibili, YouTube, and Xiaohongshu.
- Not a fit if
- Users unwilling to install a browser extension and log into multiple platforms
- Users who need a polished Safari or Firefox experience
- Source review
- 63/100 · Some gaps
What does this agent do, and when should you use it?
OpenBiliClaw is an open-source, local-first, cross-platform content discovery agent that builds a deep psychological user profile from your behavior and actively searches for content across multiple platforms (e.g., Bilibili, Reddit, YouTube, Zhihu). Its core differentiator is a five-layer 'soul portrait' that infers MBTI, cognitive style, and deep needs, enabling proactive exploration rather than passive matching. The project includes a browser extension, desktop and mobile web UIs, and a Python backend with local SQLite storage. Installation options range from desktop installers (macOS/Windows) to AI-assisted one-line deployment or Docker. It supports multiple LLM providers (OpenAI, Claude, Gemini, Ollama) and offers an Agent Bridge for integration with OpenClaw, Claude Code, and similar hosts, enabling proactive pushes and Socratic dialogues. Privacy is emphasized: all data stays local by default, with optional export and migration.
OpenBiliClaw operates as a self-hosted local agent: On first run, openbiliclaw init collects opt-in signals from logged-in platforms via adapters like bilibili_adapter and xiaohongshu_adapter, feeding data to an LLM to generate a five-layer soul portrait (MBTI, cognitive style, deep needs) stored in SQLite. Then, using the discover command or a background daemon, it proactively searches across platforms (e.g., discover --source douyin) for content matching the portrait. Candidates are pooled, evaluated by a shared evaluator with diversity and deduplication, and presented via the web interface (/web client and /m mobile), browser extension sidebar, or Agent Bridge. The system explains recommendations like a friend, and user feedback (like/dislike/chat) is immediately incorporated into the portrait. It also supports optional visual embeddings and keyframe analysis for richer recommendations.
- Content enthusiasts who want to escape platform algorithms and discover novel, cross-domain content aligned with deep interests across Bilibili, YouTube, and Xiaohongshu.
- Researchers or analysts needing centralized discovery and discussion from niche communities like Zhihu, V2EX, and Linux.do.
- Privacy-conscious users who want a recommendation system that runs entirely locally, keeping all behavioral data on their own devices.
- Technical users who prefer CLI or AI-assistant-driven deployment and deep customization of the recommendation engine.
- Developers integrating with agent hosts like OpenClaw, Claude Code, or Codex to enable proactive content pushes and intelligent conversations.
- Creators or marketers analyzing trends and audience interests across multiple platforms to inform content strategy.
How do you install or deploy this agent?
- Install the browser extension: one-click from the [Chrome Web Store](https://chromewebstore.google.com/detail/cdfjfkdjjhdaccbldipkjhpibnfbiamg) or manually from the [Latest Release](https://github.com/whiteguo233/OpenBiliClaw/releases/latest) as a zip.
- Deploy the backend:
- Option A (Desktop installer): download the macOS .dmg or Windows .exe installer from the Latest Release; it runs in the menu bar/tray.
- Option B (AI one-line deployment): paste the following into Claude Code, Codex CLI, or Cursor:
Please deploy the OpenBiliClaw backend following the instructions at https://raw.githubusercontent.com/whiteguo233/OpenBiliClaw/main/docs/agent-install.md (Use Bash's curl to download the doc, not WebFetch — it will lose critical instructions).
- Configure content sources: log in to at least one supported platform (default Bilibili) in the browser with the extension.
- Access UIs: after starting the backend, visit
http://127.0.0.1:8420/webfor desktop or use the mobile QR code for/m/.
How do you use this agent?
After installation, use the following commands or UI actions:
- Initialize: run openbiliclaw init and follow the wizard to select sources and LLM, generating the portrait and first discovery.
- Start backend: openbiliclaw start to launch the service and web UIs.
- Manual discovery: openbiliclaw discover (or openbiliclaw discover --source douyin for a specific platform).
- View recommendations: openbiliclaw recommend
- View persona: openbiliclaw profile
- Interact: browse recommendations on the web UI, provide feedback (like/dislike/chat) to train the agent.
- Agent Bridge: mount the repo to OpenClaw or similar host and follow the docs to enable proactive pushes and conversations.
What are this agent's strengths and limitations?
- Genuinely local-first: all data (SQLite, config, profile) stays on your machine by default; no cloud account means unrivaled privacy.
- Proactive exploration: uses 'interest probes' to guess and discover new domains you might like, actively breaking filter bubbles.
- Cross-platform integration: supports 12+ mainstream platforms with unified profiling and recommendations, bridging silos.
- Highly customizable: multiple LLM providers (OpenAI, Claude, Gemini, etc.) plus local Ollama, with per-module instance chains and failover.
- Agent-friendly: ships an Agent Bridge and skill, integrating seamlessly with OpenClaw, Claude Code, and others.
- Flexible deployment: desktop installers, AI one-liner, Docker, and script options for all skill levels.
- Setup complexity: requires installing both extension and backend, and logging into multiple platforms; there's a learning curve.
- External LLM dependency: high-quality recommendations typically need a cloud LLM API (cost) or a large local model download.
- Desktop installers are experimental: unsigned macOS packages and Windows SmartScreen prompts add friction.
- Browser compatibility: primarily Chrome-based; Firefox support is improving but not fully polished; Safari not mentioned.
- Some features need the extension online: platform session sync and certain task execution are tied to it.
How does this agent compare with similar options?
The README compares OpenBiliClaw to platform-native recommendation systems and keyword-filter extensions. Native recommendations rely on collaborative filtering within a single platform, keyword filters use tag matching, and OpenBiliClaw uses a psychological profile with five-layer memory. Native and keyword tools operate on a single content source, while OpenBiliClaw spans multiple platforms. On filter bubbles, native narrows, keyword doesn't address it, OpenBiliClaw deliberately probes for new interests. Data ownership: native belongs to the platform, keyword tools are typically cloud, OpenBiliClaw is 100% local. Recommendations from native are 'guess you like', keyword none, OpenBiliClaw explains like a friend. Customizability: native none, keyword low, OpenBiliClaw high (switch LLM, edit profile, write skills).
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OpenBiliClaw: Local-First Cross-Platform AI Content Discovery Agent This agent | 63 · Some gaps | Browser extensionFree + model costs | ★ 3.5k | 2d ago | Python | ChatGPT · Codex · Claude Code |
| Rowboat | 43 · Major gaps | Desktop appFree + model costs | ★ 18k | 3d ago | TypeScript | Codex · Claude Code |
| COMPASS Skills | 85 · Good | CLIFree | ★ 758 | today | Python | Codex · Claude Code |
| Lotti Private Logbook | 80 · Good | Desktop appFree + model costs | ★ 1.2k | today | Dart | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: extension defaults to local backend permissions, remote requires user grant; initialization requires explicit consent for personal data; data flow described clearly (local SQLite, cookie sync, public discovery); sensitive data handling documented (no passwords saved, cookie values not uploaded); dependency security has basic checks (CI, Ruff, MyPy) but no vulnerability scanning found; external effects controlled (read-only imports, write-back requires authorization); rollback limited (config backup, data not deleted); source attribution clear (per-platform login, data usage). Deductions: no dependency vulnerability scanning, incomplete rollback.
Evidence shows: README and code structure consistent, version numbers match; dependencies declared, CI covers tests; failure messages designed (e.g., uninitialized hints, health checks). Deductions: static review cannot verify runtime reliability, but no obvious contradictions.
Evidence shows: clear target users (general, developers, AI assistants), rich scenarios (desktop, mobile, extension); capability boundaries clear (per-platform support, login requirements); trigger mechanisms clear (commands, events); environment fit good (cross-platform, Docker, HTTPS). Deductions: some advanced features rely on external services (e.g., Ollama), but documented.
Evidence shows: clear information architecture (README, docs, code structure); detailed install instructions (multiple methods); stable naming (versions, commands); examples and FAQ exist (docs, discussion threads); known limitations mentioned (unsigned, experimental); license clear (MIT); changelog exists; maintenance responsibility clear (community, GitHub). Deductions: some docs in Chinese may affect non-Chinese users.
Evidence shows: usable outputs (recommendations, profile, chat); high marginal value (cross-platform, proactive recommendations); reasonable cost-benefit (local, bundled models). Deductions: static review cannot verify actual effectiveness, but design is sound.
Evidence shows: claims supported by docs (e.g., features correspond to code); cross-source corroboration limited (only internal tests); fact/inference separation clear (profile inference explained). Deductions: lack of independent verification, but internal test coverage broad.
- Publisher identity unverified; proceed with caution.
- No dependency vulnerability scanning found; recommend self-check.
- Rollback mechanism incomplete; back up data before upgrades.
- Some documentation in Chinese may affect non-Chinese users.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
desktop-v0.3.229 Pre-release
Release notes excerpt
桌面预发布安装包。macOS 安装包均经过 Developer ID 签名和 Apple 公证；缺少凭据或公证失败时停止对应构建，不生成实验包。后端源码仍为 source-only，此处只发桌面应用二进制。 安装 - macOS：退出旧版，打开 DMG，将 OpenBiliClaw 拖到 Applications 后启动。无需解除隔离；首次打开可能显示正常的互联网下载确认。 - Windows：SmartScreen 弹窗点「更多信息 → 仍要运行」。 自带本地 Ollama + bge-m3 embedding，开箱即用，常驻 macOS 菜单栏 / Windows 系统托盘。数据在 ~/OpenBiliClaw（macOS / Linux）/ %USERPROFILE%\OpenBiliClaw（Windows），升级 / 卸载不动它。Intel mac（x64）.dmg 未自动构建，需要时用手动的 build-installers.yml 产出后补挂。 What's Changed …View on GitHub ↗ -
Release notes excerpt
What's Changed * docs: record v0.3.228 release and store receipts by @whiteguo233 in https://github.com/whiteguo233/OpenBiliClaw/pull/294 * feat: add TikTok discovery via signed Web API by @whiteguo233 in https://github.com/whiteguo233/OpenBiliClaw/pull/275 * feat: add TikTok account bootstrap and opt-in incremental sync by @whiteguo233 in https://github.com/whiteguo233/OpenBiliClaw/pull/286 * feat: 新增可选 Parallel MCP 关键词 grounding 后端 by @georgeatparallel in https://github.com/whiteguo233/OpenBiliClaw/pull/287 …View on GitHub ↗ -
openbiliclaw-v0.3.229 OpenBiliClaw v0.3.229
Release notes excerpt
This is the user-facing aggregate release. It keeps the current backend source tag, browser extension packages, and desktop installers visible together. Current Channels - Backend source: [backend-v0.3.229](https://github.com/whiteguo233/OpenBiliClaw/tree/backend-v0.3.229) - Browser extension: [extension-v0.3.229](https://github.com/whiteguo233/OpenBiliClaw/releases/tag/extension-v0.3.229) - Desktop installer: [desktop-v0.3.229](https://github.com/whiteguo233/OpenBiliClaw/releases/tag/desktop-v0.3.229). …View on GitHub ↗