Skip to content

README.md

Latest commit

 

History

History
125 lines (86 loc) · 4.6 KB

README.md

File metadata and controls

125 lines (86 loc) · 4.6 KB
Raw

Sirchmunk OpenClaw Skill

Use Sirchmunk as an OpenClaw skill so that any OpenClaw-compatible AI agent can search your local files with natural language — no embedding database, no indexing, no ETL.

Published on ClawHub: https://clawhub.ai/wangxingjun778/sirchmunk

Quick Start

1. Install Sirchmunk on the machine that will run the server

pip install sirchmunk
sirchmunk init          # creates ~/.sirchmunk/.env

Edit ~/.sirchmunk/.env — at minimum set:

LLM_API_KEY=sk-...
LLM_BASE_URL=https://api.openai.com/v1   # or any OpenAI-compatible endpoint
LLM_MODEL_NAME=gpt-4o

# Optional: default search directories (comma-separated)
SIRCHMUNK_SEARCH_PATHS=/path/to/your/docs,/another/path

2. Start the Sirchmunk API server

sirchmunk serve          # default: http://0.0.0.0:8584

Verify it is running:

curl http://localhost:8584/api/v1/search/status

3. Install the skill in OpenClaw

Install the skill from ClawHub:

npx clawhub@latest install sirchmunk

Or copy the sirchmunk/ directory from this repo into ~/.openclaw/skills/sirchmunk/.

4. Use it

The agent can now invoke the sirchmunk_search tool. You can also call the script directly:

# Use default SIRCHMUNK_SEARCH_PATHS from the server environment
~/.openclaw/skills/sirchmunk/scripts/sirchmunk_search.sh "What is the reward function?"

# Override with an explicit path
~/.openclaw/skills/sirchmunk/scripts/sirchmunk_search.sh "auth flow" "/path/to/project"

Usage examples / 使用示例

Below are screenshots of local RAG-style retrieval with the Sirchmunk skill inside OpenClaw TUI (openclaw-tui): the agent calls sirchmunk to read matching rows from local CSVs (e.g. OBQA / science QA data), then summarizes in the requested language.

English — local search example

Query (illustrative): ask about ancient societies and light reflection; answer grounded in a local file such as obqa.csv, attributed as “Found using sirchmunk local search”.

English OpenClaw TUI: Sirchmunk local file search with markdown answer

中文 — 本地检索示例

示例问题（示意）： 关于自制冰棍是否“做好”时温度计应显示的读数；代理用 sirchmunk 检索本地 CSV（如 obqa.csv、arc_easy.csv 等），并用中文结构化回答。

中文 OpenClaw TUI：Sirchmunk 本地文件检索与中文回复

How It Works

The skill is a thin HTTP client. When invoked it sends a POST request to the local Sirchmunk server:

curl -s -X POST "http://localhost:8584/api/v1/search" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "your question",
    "mode": "DEEP",
    "response_format": "rich"
  }'
  • paths is optional in the request body. If omitted, the server uses SIRCHMUNK_SEARCH_PATHS from its environment, then falls back to its working directory.
  • mode can be DEEP (default, comprehensive, 10-30 s), FAST (2-5 s), or FILENAME_ONLY (no LLM).
  • response_format can be rich (default Markdown evidence report), minimal, context, or json.

See the main HTTP Client Access (Search API) section for the full parameter reference, SSE streaming endpoint, and client examples in Python / JavaScript.

File Structure

recipes/openclaw_skills/
├── README.md
├── assets/
│   ├── example1.png                # Chinese usage screenshot
│   └── example2.png                # English usage screenshot
└── sirchmunk/
    ├── SKILL.md                    # OpenClaw skill manifest
    └── scripts/
        └── sirchmunk_search.sh     # Wrapper script invoked by the agent

Security Notes

  • The Sirchmunk server runs locally and binds to localhost by default. File contents are sent to the configured LLM endpoint for analysis. If you use a cloud LLM, be aware that searched content leaves the machine.
  • Prefer a local LLM (Ollama, vLLM, etc.) or restrict SIRCHMUNK_SEARCH_PATHS to non-sensitive directories if data exfiltration is a concern.
  • Do not expose port 8584 to untrusted networks without additional authentication.

Links