discord-rag

Discord RAG Skill

Overview

Retrieval-Augmented Generation over Discord message history. Ingest channel messages into a local LanceDB vector database, then search semantically — find messages about a topic even when the exact words don't match.

How It Works

1. rag_ingest(channel_id, limit=50) fetches recent messages, generates embeddings via sentence-transformers (all-MiniLM-L6-v2), and stores them in LanceDB at data/discord_lancedb.
2. rag_query(query_text, top_k=10) finds the top_k most semantically similar messages and returns them with channel name, author, and timestamp.

Data persists across restarts. You can use different table_name values to organize data from different channels or servers.

Workflow

1. Ingest Messages

discord(operation="rag_ingest", channel_id="...", limit=100)

Start with a reasonable limit (50-100 messages). Larger ingests take longer due to embedding generation.

2. Search

discord(operation="rag_query", query_text="discussion about deployment", top_k=10)

Prefer narrow, specific queries for best results.

3. Act on Results

The returned messages include channel_id, author, timestamp, and the message content. From here you can:

• Use get_messages to fetch the full thread context
• Use send_message to respond in the same channel
• Use export_messages to archive the discussion

Tips

• Ingest separate channels into separate tables by setting table_name="..." — this keeps contexts clean.
• The first ingest call loads the embedding model (takes ~5-10s). Subsequent queries are fast.
• RAG requires sentence-transformers and its dependencies (installed via uv sync --extra rag).
• Data is local-only. No external API calls are made during ingest or query.

Env Dependencies

• LANCEDB_DISCORD_PATH — Override LanceDB storage path (default: data/discord_lancedb).