Introducing IntentiAI: The Intent Classification Engine for AI Agents Hey everyone! I built IntentiAI because AI agents today are fundamentally broken at understanding what users want. Most AI applications are intent-blind. They treat every query the same, sending everything to one model regardless of what the user is actually asking. That's not AI. That's just a fancy if-else statement. The real problem: Without understanding intent, you can't: - Route queries to the right model - Build guardrails against out-of-scope requests - Understand what users actually want - Improve your AI over time My solution: IntentiAI is an intent classification engine that: - Achieves 91% F1 accuracy on BANKING77 (77 fine-grained intents) - Classifies in <20ms on CPU (no GPU required) - Detects out-of-scope queries with 94% recall Once we know what the user wants, intelligent routing becomes trivial. Cost savings? That's just the byproduct of matching query complexity to model capability. Why open source the frontend + widget? Because I know people are skeptical about giving API keys to a new service. The widget and frontend are open source so you can verify exactly how data flows through our system, no hidden tricks, no data collection. The core classification engine stays private (that's our competitive edge), but the parts that matter for trust? Completely transparent. Would love your feedback! What would make intent classification more useful for your AI agents?

Intenti AI

Intent Classification: 91% Accuracy, 70% Lower Costs

Intent Classification: 91% Accuracy, 70% Lower Costs