How dedup works
After each completed session, Celina embeds the question with all-MiniLM-L6-v2 (384-dim unit vector). The next time a question arrives, Celina computes cosine similarity against all active memories. Similarity ≥ 0.85 triggers a cache hit and skips all paid x402 calls. If the model is unavailable, Celina falls back to address-based matching (shared 0x address = 0.9 similarity).
| Question | Verdict | Conf. | Embed | Created | Expires |
|---|---|---|---|---|---|
Deep-dive 0x779ded0c9e1022225f8e0630b35a9b54be713736 (deep-dive triggers the sub... 0x779ded… | The token 0x779ded0c9e1022225f8e0630b35a9b54be713736 is a re... | 90% | 384-dim | 4h ago | 20h |