We Built a Production Agent (and Open-Sourced Everything We Learned)

Tiger Data's recent endeavor focused on developing an AI-native application called Eon, a Slack-native assistant designed to convert institutional knowledge into instant answers by leveraging conversational memory, context, and reliability. The project, which became widely used within the company in just six weeks, involved tackling three primary challenges: building memory that understands time using TimescaleDB, providing context without noise through Model Context Protocol (MCP) servers for platforms like GitHub and Linear, and ensuring reliability with features like durable event processing and automatic retries. The team open-sourced the components developed, including tiger-agents-for-work for Slack agent frameworks, and MCP servers for GitHub, Linear, and documentation search, allowing others to replicate or innovate upon their work. This initiative highlights the importance of durable event handling and focused tools for AI agents, showcasing the potential of using Postgres infrastructure for AI applications and marking the beginning of a broader journey into developing agentic capabilities.