Setoku: A Self-Hosted Knowledge Server for AI Agents Built on Claude
What Is This Product?
Setoku is a self-hosted knowledge server designed to act as a central brain for AI agents. Developed at Hedgy, it ingests data into a ClickHouse data lake and provides a lightweight knowledge infrastructure that stores contextual information about that data—like how to count an active user or which column to check for lifetime value. The system is exposed to AI agents via the Model Context Protocol (MCP), allowing agents to query and record insights as they work.
Problem It Solves
AI agents often struggle with fragmented or poorly structured data. They need not just raw data but also the contextual knowledge around it—rules of thumb, business logic, and gotchas. Setoku addresses this by offering a single, self-hosted service where data and its associated knowledge live together. Agents can access the data lake and the knowledge base through a unified interface, reducing errors and improving decision-making without relying on external inference APIs.
How AI Works
Setoku does not perform AI inference itself. Instead, it leverages Claude subscriptions (via API) to power the agent's reasoning. The MCP protocol enables Claude to interact with Setoku: reading data, recording new insights, and retrieving relevant knowledge. The system encourages agents to document gotchas and findings as they discover them, creating a growing knowledge base. A minimal admin interface allows humans to audit and prune that knowledge, ensuring quality and relevance over time.
Business Model
Setoku is currently a side project used internally at Hedgy, not a commercial product. There's no pricing or revenue model disclosed. The operating cost appears to be limited to the underlying Claude subscriptions and the infrastructure for hosting ClickHouse and the Setoku server. Because it's self-hosted, users maintain control over data and costs. The open-source nature (likely, given it's on GitHub) means others can adopt it freely, but official support or cloud hosting could be future monetization paths.
Growth Opportunity
The rise of AI agents in enterprise and development contexts creates demand for infrastructure that connects agents to proprietary data. Setoku fills a gap between raw data lakes and agent-friendly knowledge retrieval. By being self-hosted and MCP-compatible, it appeals to teams that prioritize data sovereignty and want to avoid vendor lock-in. If the project gains traction on GitHub and Hacker News, it could evolve into a paid offering or attract sponsorship. The main growth lever is clear documentation and success stories from early adopters.
Key Takeaways
• Setoku is a practical tool for teams building AI agents that need structured access to company data and business context.
• It relies on Claude for intelligence and ClickHouse for storage, with MCP as the glue.
• No inference runs on Setoku itself—it's purely a data and knowledge layer.
• The project is early stage: no revenue model, but low operational cost makes it viable as a side project.
• For indie hackers and AI builders, Setoku demonstrates a lean approach to giving agents long-term memory and domain-specific knowledge without expensive infrastructure.
