A team at startup Hedgy has open-sourced Setoku, a self-hosted knowledge server designed specifically for AI agents. The project landed on Hacker News as a "Show HN" post on July 23rd, 2026, scoring modest visibility but catching the attention of developers building agentic workflows that need persistent, queryable memory.

What Is Setoku?

Setoku describes itself as a self-hosted company brain. The architecture centers on a ClickHouse data lake for ingesting and querying structured information, paired with what appears to be lightweight knowledge infrastructure on top. Crucially, the system is powered entirely by Claude subscriptions—meaning you're plugging in your own Anthropic API keys rather than trusting a third-party service with your proprietary data.

Why Self-Hosting Matters for AI Agents

The pitch is straightforward: as AI agents become more capable and start handling sensitive business logic, sending that context to external APIs becomes a liability. Setoku lets teams keep their knowledge bases local while still giving autonomous agents the ability to retrieve relevant information at runtime. It's an architectural pattern that's gaining traction in enterprise AI circles—think retrieval-augmented generation (RAG) stacks but purpose-built for multi-agent systems.

The Technical Stack

Based on the HN post, Setoku uses ClickHouse as its data layer—a columnar database known for analytical query performance at scale. This makes sense for a knowledge server that might field complex queries across large document corpora. The Claude integration suggests semantic search and summarization capabilities, though full technical documentation wasn't available in the source material.

The Hedgy Connection

The project emerged from internal iteration at Hedgy, suggesting it solved a real problem for the team before they decided to share it publicly. This is classic hacker culture—build tools for your own pain points, then open-source them when you think others might benefit. The low HN score (6) indicates it's early days, but side projects like this often find their audience through word of mouth in specialized communities.

Key Takeaways

  • Setoku provides a self-hosted knowledge server architecture for AI agents, avoiding reliance on third-party data services
  • ClickHouse powers the data ingestion and querying layer, handling structured knowledge at scale
  • Claude subscriptions drive the AI capabilities, giving agents semantic search and summarization over private corpora
  • The project targets teams building agentic systems that need persistent, queryable memory without sacrificing data sovereignty

The Bottom Line

Setoku is a niche play for teams running multi-agent architectures who want to keep their knowledge bases in-house. It's not going to replace your existing RAG pipeline overnight, but if you're operating Claude-powered agents in sensitive environments, this architecture pattern is worth studying. Watch this space—tools like this are early signals of where enterprise AI infrastructure is heading.