Anchor sentence
In SignalPipe, an anchor sentence is a short, buyer-voice example of the intent a product wants to detect — embedded with text-embedding-3-small and used as the cosine-similarity reference for every incoming post.
Each product configures a small set of anchor sentences (typically 5–10) describing the moment of buying intent its ideal customer might express. Example: "I need a tool to monitor Reddit and Hacker News for sales leads." These are written in the buyer's voice, not the seller's — anchors describing your features instead of the buyer's pain produce noisy matches.
Mantidae embeds each anchor with OpenAI text-embedding-3-small at product-load time and caches the vectors in memory. Incoming RSS / Reddit / HN posts are embedded the same way. The cosine similarity to the closest anchor becomes the embedding component of the multi-factor signal score.
Anchors can be hot-reloaded without restarting the API — call POST /products/reload after editing the products table.
Related terms
- Signal score →Final 0–100 score on a SignalPipe lead, after multi-factor scoring, competitor floor, and RL weight.
- Station →An RSS, Reddit, or Hacker News feed configured to listen for buying signals for one SignalPipe product.
- Role-aware swarm →3 LLM judges × 3 roles = 9-prompt matrix that calibrates SignalPipe drafts to the strength of the signal.
Cite as: https://signalpipe.io/glossary/anchor-sentence