moss-docs

What This Skill Does

The moss-docs skill provides OpenClaw agents with comprehensive capabilities to interface with the Moss semantic search runtime. It acts as an abstraction layer for managing real-time retrieval-augmented generation (RAG) pipelines, enabling agents to store, index, and query information with sub-10ms latency. By leveraging this skill, an agent can autonomously manage document lifecycles, perform complex hybrid searches, and switch between embedding models like moss-minilm and moss-mediumlm based on the specific accuracy or performance needs of the task. Whether you are building an offline-first assistant or a cloud-synced documentation platform, this skill manages the vectorization and search logic so you don't have to.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal or agent console:

clawhub install openclaw/skills/skills/coderomaster/moss

Ensure your environment has the necessary network permissions to reach the Moss API endpoint at https://service.usemoss.dev/v1 if you intend to utilize cloud-sync features.

Use Cases

1. Personal Knowledge Base: Create a private semantic index of local markdown files or research papers to allow the agent to answer questions based on your specific document library.
2. Dynamic Technical Support: Use the agent to ingest API documentation, enabling instant retrieval of specific technical patterns during a live coding or debugging session.
3. Hybrid Content Discovery: Combine keyword-based filtering for exact technical terms with semantic vector search to find relevant context in messy, unstructured datasets.
4. Context-Aware Memory: Use the indexing features to store agent-specific conversation summaries, allowing the model to recall long-term facts across multiple sessions.

Example Prompts

1. "Initialize a new index named 'project-docs' using the moss-mediumlm model and add all files from my current project directory to it."
2. "Perform a hybrid search in the 'technical-manual' index for 'how to configure load balancer timeouts' and provide the top 3 relevant sections."
3. "List all active indexes and verify the document count for the 'daily-logs' repository."

Tips & Limitations

• Model Selection: Use moss-minilm for scenarios requiring extreme speed on edge hardware. Reserve moss-mediumlm for tasks where high semantic accuracy and nuanced query matching are critical.
• Search Tuning: When using the Python SDK for hybrid search, spend time calibrating the alpha weighting. A higher alpha favors keyword matching (BM25), while a lower alpha relies more heavily on vector semantic similarity.
• Data Privacy: If running in an agent context with sensitive info, ensure you use the on-device index storage rather than cloud-synced repositories to keep data strictly local.
• Batch Processing: To optimize performance, always add documents in batches using the add_docs() method rather than individual calls to minimize network overhead.