+++ type: home +++

Online Demo

Try Familiars right now — no local install needed. The demo runs on a shared cloud instance: you can talk to a real agent, create todos / schedules, and try file generation and the skill workshops.

在线演示

🔄 检测中

✅ 会话✅ 待办/日程✅ 文件生成✅ CG动画工坊✅ 技能工坊·⛔ 记忆⛔ 浏览器⛔ 本机命令⛔ OS输入⛔ 电话/社交

🎮 进入演示 ↗

Are these "fake buttons"?

No. The demo runs a CPU-only full bridge (the src/ FastAPI backend + the VITE_DEMO_MODE web build of the Electron renderer SPA). Chat goes through a real cloud LLM (demo accounts are billed for real), and tools really execute inside an isolated cloud sandbox per visitor.

• Isolation: the demo process uses a separate WORKSPACE_ROUTE + FAMILIARS_DB_PATH, with zero overlap with production / your local data;
• Sandbox: every visitor gets their own sandbox directory (sessions/<account>/workspace); all file reads/writes are anchored inside their own sandbox — it cannot read or write your computer, other visitors' sandboxes, or any other path on the server;
• Two layers of protection: dangerous tools (code execution, OS input, browser/network, phone/social, etc.) are removed from the agent's toolset entirely, and a runtime interception layer backs this up;
• Ephemeral: every file written to the sandbox is auto-deleted after 3 hours — the demo is a temporary experience; download anything important.

What you can do in the demo

Capability	Description
Real chat	A real manager agent (agent 0), cloud LLM + real tool execution
Todos / schedules / reminders	Say "add a todo for tomorrow 3pm" — the agent really stores it and can look it up
File generation	Word / PPT / Excel skill chains work fully; outputs land in the sandbox and are pasted back into the chat as attachments (no download links needed)
File upload	Upload attachments into your own sandbox and reference them in chat
CG animation workshop	Character CG asset generation / preview / compositing (skill-driven)
Skill workshop	Skill editor (write / validate / test skills)
Image processing	Lightweight in-sandbox image ops such as grid overlay / keyframe extraction
Settings / About	View runtime status and project intro

What the demo cannot do

Compared with the desktop client, the demo disables these capabilities (tools are removed from the agent's tool surface; UI items are greyed out and marked "not available in the demo"):

Capability	Demo	Desktop
Agent chat (real cloud LLM)	✅	✅
Todos / schedules / reminders	✅ (chat-driven)	✅ (incl. standalone app cards)
File generation (Word / PPT / Excel)	✅ (sandboxed, 3h TTL)	✅
CG animation / skill workshop	✅	✅
Creative workshop (Recipe / Flow editor)	❌ (desktop cards greyed out)	✅
Long-term memory (RAG)	❌ fully off (no retrieve / write / explore)	✅
Tool approval popups	❌ approval gate off, tools run directly	✅ gate on by default
Browser automation / web search	❌ (browser_*, web_search/fetch removed)	✅
Code / process execution	❌ (shell, python/node, pip/npm/git, subprocess, open apps, system queries)	✅
OS input control / screenshots	❌ (mouse/keyboard/clipboard/windows, screenshots/recording)	✅
Phone / social / email	❌ (dial, call sessions, send email, social platform posting)	✅
Network egress / SSH	❌ (http_request, ssh/scp)	✅
OCR text recognition	❌ (paddleocr not installed on the demo box)	✅
Digital human / voice / local models	❌ (no GPU on the VPS: no TTS, ASR, image generation, QLoRA/SoVITS)	✅
Create agents / edit agent settings	❌ (UI disabled)	✅

Only About / Settings / CG animation workshop / Skill workshop open from the desktop app panel; the other apps (creative workshop recipe editor, reports, todo cards, message listeners, app store, skill approvals, guild/Ventaris) are shown but marked "not available in the demo".

Want all these capabilities? Install the desktop client — your data stays on your machine and models/tools are under your control. (Install guide)

Fair-use rules (seats and quota)

The demo runs on a shared cloud instance and every real chat consumes cloud LLM cost, so please follow these rules to keep it fair:

① Concurrent seats (15)

• At most 15 active sessions are admitted at the same time;
• When full, you'll see "currently full, please retry later" — seats free up automatically when there is room;
• At most 2 seats per network (IP) so one person cannot monopolize all seats;
• Sessions idle for 30 minutes release their seat automatically;
• The status card at the top of the page shows live "N/15 seats remaining".

② ¥20 quota per account

• Each browser fingerprint = one account (da_*) with a ¥20 (2000 credits) quota, deducted by cumulative LLM cost;
• Revisiting with the same browser keeps the same account; the balance persists across sessions; spending stops when exhausted;
• Please don't clear your browser fingerprint repeatedly to reclaim quota — it breaks demo cost control and the server rate-limits by network.

③ File lifecycle

• All sandbox user files (uploads, generated, temporary) are auto-deleted after 3 hours;
• Sandboxes of accounts unused for 7 days are cleaned up entirely; quota records are kept for 30 days;
• Download anything you want to keep.

④ Privacy boundary

• Do not enter real sensitive information in the demo (shared instance; content may be used for demonstration);
• Each visit generates a local temporary identity (browser fingerprint) to attribute sessions and quota;
• The demo has memory disabled — agents don't remember your visits and nothing is written to long-term memory.

Demo walkthrough

1. Opening: say "hi, introduce yourself" to agent 0;
2. Todo: say "add a todo for tomorrow 3pm to hand in the demo report", then ask "what todo did I just create?" — the agent answers within the same conversation;
3. File generation: say "generate a Word doc with a bounty-hunter character intro" — the document comes back as an attachment you can open directly;
4. Skill workshop: open the skill workshop app and try the skill editor — writing, validating and testing are real;
5. Schedule reminder: say "remind me to drink water at 9am every day" — the timed reminder really fires.

To experience memory, approvals, browser, and the full feature set, use the desktop client.

Deployment & ops

The demo is deployed independently on a VPS (CPU-only, no local models). Full setup: see the VPS deployment guide — covering SPA build, bridge startup, reverse proxy with SSE, per-user sandbox isolation and file lifecycle cleanup.

Deploy note: POST /api/demo/token serves the demo-only process (sandbox workspace). Do not expose this endpoint on a production bridge — that would hand arbitrary visitors full access to the current workspace.

Feedback

Hit a problem in the demo? Two ways: submit feedback from the page itself (see Feedback), or open a GitHub Issue.