Skills Data and integration
Vectorising support conversations
Semantic search over support history is genuinely useful: finding the precedent for an unusual case, grounding an AI agent, or locating every conversation about a topic without guessing keywords.
npx rulebase-skills install cx-conversation-embedding-pipelineWhen to use it
Reach for this when someone says any of these — they are the phrases the skill itself triggers on:
- “make our support conversations searchable”
- “build a vector index over tickets”
- “semantic search across transcripts”
How it works
The method, in the order the skill runs it. The full procedure — tables, worked examples and the edge cases — is in the skill itself.
Chunk on turn boundaries, not on character counts
Fixed-size chunking is the default in most tooling and it is wrong here. Splitting mid-turn produces chunks containing half of the customer's problem and half of the agent's answer, which retrieve badly and read worse.
Attach the metadata that makes retrieval usable
An embedding index over conversations with no filters is much less useful than it looks — almost every real query is scoped.
Do the PII work before embedding, not after
The decisions, in order.
Deletion has to reach the index
The failure that will find you later: Test the deletion path.
Evaluate retrieval, or you will not know it is bad
An embedding index always returns something, which is why bad ones survive.
Related skills
Free and open source, and vendor-neutral — it reads the conversations from whichever helpdesk you already run. Browse all 149 skills · connect your helpdesk over MCP · source on GitHub
