Ganu.

Ganu is an AI customer support bot for technical products, designed around a deliberately narrow thesis: the weakest link in customer support is the handoff from bot to human, and that's exactly where every existing AI support tool fails. Most bots dump a raw chat transcript on the support engineer, who has to read the entire conversation to understand what's going on — killing productivity at the moment the customer most needs speed.

Ganu fixes the handoff with two new components: (1) a Silent Evaluator that runs alongside every conversation, continuously assessing resolution confidence, sentiment, and escalation criteria — deciding when escalation truly makes sense (not too early, not too late); (2) a Structured Handoff Card displayed in a side panel for the human agent: Concise Issue Summary, Steps Already Tried, Recommended First Action, Customer Sentiment, Routing Suggestion. The next release roadmap extends this with automatic Salesforce case creation populated from the handoff card and auto-routing to the appropriate team.

Multi-agent architecture, each agent matched to its role: Opus 4.6 for the Support Bot (capability matters more than cost when the customer is talking to it directly), a separate Evaluator agent running every turn (straightforward task, JSON output, candidate for an optimal cost-efficient model), and a Handoff Generator that produces the structured card on escalation. Required inputs across the three: full conversation transcript, escalation criteria + confidence thresholds, support-bot product feature scope.

Built for product companies with technical customers — primary users are the customer's data engineers and engineering staff using the bot to resolve product issues; secondary internal users are Tier 1 Customer Support Engineers receiving escalations. B2B model — the customer is the product company; the bot's users are *their* customers. Differentiator: open unified data platform with the silent evaluator and structured handoff as the trust layer that solves the escalation precision problem every existing support bot under-serves.