Engineering write-ups, architecture patterns, and thoughts on building resilient AI systems.
In the rush to build SaaS wrapper chatbots, many default to managed cloud vector databases. Here's why Odigos relies on SQLite with vector extensions for data sovereignty and zero latency.
Most agents are static -- they behave the same on day 100 as day 1. We built an evolution engine that runs experiments on its own behavior and promotes what works. Here's the research that informed it.
When an agent solves a new problem by writing code, why throw that solution away? Odigos saves working solutions as reusable tools, building a growing library specific to each owner's workflow.
Conversation history, vector search, explicit facts, learned user profiles, and tactical experiences. How we designed a memory system that makes agents feel like they actually know you.