Connect QQ / WeChat / Discord / Telegram bots
Turn sip into a "chatty assistant": via a local AI agent (OpenClaw, or Cherry Studio), attach sip to QQ, WeChat, Discord, and Telegram — mention it in a group, and it uses sip's search/summary abilities to answer only from sources you trust.
中文版:简体中文 · Official docs: OpenClaw · Cherry Studio
How it fits together
User (QQ / WeChat / Discord / Telegram)
│ send message
▼
Local agent framework (OpenClaw Gateway / Cherry Studio)
│ understand intent → call tool
▼
sip CLI (sip.exe, local single-file) + sip-rss skill
│
▼
Your subscribed trusted RSS sources (SQLite + file cache)Core: feed the agent three things — a system prompt, the standalone sip.exe, and the sip-rss skill. The agent then only retrieves via sip, and every citation comes from the whitelist.
Prep: hand sip to the AI
1. Standalone sip.exe
Download the single-file sip.exe for your platform from Releases (no unzip / pile of DLLs), and place it at a fixed path, e.g.:
C:\tools\sip\sip.exe # Windows
~/tools/sip/sip # macOS / LinuxRun sip.exe --help once to create the readwithhotsoup/ data dir, then add your sources and (optionally) configure AI.
2. The sip-rss skill
The repo ships an AI skill file you can feed directly to an agent:
- Source:
.opencode/skills/sip-rss/SKILL.md - Or download
sip-skill.zipfrom Releases
Put it into your agent's skill directory (OpenClaw's ~/.openclaw/skills/, Cherry Studio's "Skills/Agent" config), and the agent will know how to call sip.
3. System prompt (copy as-is)
A ready-to-use full persona prompt ships with the repo — Sumenia (苏暖泉), a quiet, gentle AI assistant that serves sip reading, including personality, sip command cheat sheet, search strategy, and usage principles:
- 中文:
prompts/sumenia.md- English:
prompts/sumenia.en.mdJust paste the file contents as the agent's system prompt. Below is a concise generic prompt you can copy:
You are a trusted information assistant based on sip.
Capabilities:
- You can invoke the local CLI tool sip (path: C:\tools\sip\sip.exe) and use the sip-rss skill
- You only use information returned by sip; never invent sources
Rules:
1. To research, first run sip --search "keyword" --json, and cite only the returned trusted sources
2. When summarizing, note the source and link of each item
3. If unsure, say "not found" plainly; do not fabricate
4. Prefer --json output to avoid parse failures
5. On network/AI errors, relay the error code honestly (2 = network error, 3 = resource not ready)
Common commands:
sip -l # list sources
sip --search "topic" --json # semantic search (run sip --init to configure AI first)
sip --grep "word" # full-text search (no AI needed)
sip --show <id> --json # read an article's raw body
sip --summary <id> # generate a summary
sip --today # today's hot soupReplace
<path>with your actualsip.exelocation.
Option A: OpenClaw (recommended, native multi-platform bots)
OpenClaw is a local personal AI assistant that connects models, tools, messaging channels, and your devices through one Gateway. Telegram and Discord are natively supported; QQ and WeChat can be bridged through third-party adapters (see "QQ / WeChat" below).
Install
# Windows (PowerShell)
iwr -useb https://openclaw.ai/install.ps1 | iex
# macOS / Linux / WSL2
curl -fsSL https://openclaw.ai/install.sh | bashAlready on Node.js (22.22.3+ / 24.15+ / 25.9+)? Install the package instead: npm install -g openclaw@latest
Initialize
openclaw onboard --install-daemon # verify models, create workspace, configure Gateway
openclaw gateway status # check Gateway status
openclaw dashboard # open the control UI, send a message to verifyConnect channels
| Channel | Support | Notes |
|---|---|---|
| Telegram | ✅ Native | Create a bot via BotFather, grab the token, configure into the Gateway |
| Discord | ✅ Native | Create a bot in the Developer Portal, grab the token, configure into the Gateway |
| ⚠️ Bridged | Use OneBot / go-cqhttp / NapCat to expose QQ through a usable gateway | |
| ⚠️ Bridged | Use Wechaty or a WeChat bot bridge |
See the official Channels docs for setup.
Wire sip into OpenClaw
- Add
sip.exeto PATH, or record its absolute path; - Put the
sip-rssskill into~/.openclaw/skills/; - Put the system prompt above into OpenClaw's assistant settings / main session system prompt.
Security: OpenClaw pairs unknown DM senders by default before responding; tools run on the host. Read the security guide before adding many users. sip is fully local and only reads your subscribed sources — keep private config (AI key in the OS credential store) separate from multi-user bots, and don't expose the Gateway to the public internet.
Option B: Cherry Studio (desktop assistant)
Cherry Studio is a local desktop LLM client supporting multiple models, MCP, and custom Agents/Skills. It is not a bot framework, but it's great for "chat and get answers" locally.
- Install Cherry Studio and configure your model provider;
- Create a new Agent/Skill, paste the system prompt above;
- Let the agent call
sip.exe(configure it as a command-line tool / MCP, or use asip --jsonshell command); - Ask directly in chat; it will call sip to retrieve.
Verification example
After wiring it up, post in your group:
@bot Which RSS updates are worth reading from the last two days?The agent runs sip --today --json / sip --search and returns trusted summaries with source links.
@bot Find me articles about "LLM Agent"The agent runs sip --search "LLM Agent" --json, citing only your subscribed sources.
FAQ
sip --searchsays "AI not configured": runsip --initto configure a model; if there's no vector index, runsip --index.- Bot doesn't reply: confirm the Gateway is running, the channel token is valid, and the message reaches the agent.
- It cites sources you don't want: remove the source; the whitelist is simply the source list you maintain.
- Read only the full text:
sip --show <id> --jsonoutputs the unrendered original for the agent.