# komjo

> komjo maps how a software system connects — classes, methods, call relationships, tables, stored procedures, triggers, and relationships across repository boundaries — and serves that graph to developers and AI agents over MCP (Model Context Protocol). Understand your system, not just your code. A product by Pixiray (https://pixiray.ai). Currently accepting applications for the early-adopter program, which starts with a small first cohort.

The website https://komjo.ai/ is a start page with the application form, plus legal pages (imprint, privacy policy) and the confirmation pages for the mailing list. German by default with an English toggle (client-side). This file is the canonical plain-text summary in English.

## The problem

The gap between how fast a codebase grows and how fast anyone understands it is widening — from three directions:

- Scale: millions of lines grown through years of real business; no single person can hold the system in their head.
- Speed: AI assistants multiply how fast code is written; understanding does not scale the same way.
- Agents: coding agents on a large codebase grep and guess. Without structure — call graphs, blast radius, database lineage — they cannot be trusted on the systems that matter most.

## What komjo does

- Analyzes code, SQL and database schemas and builds a typed knowledge graph: classes, methods, call edges, tables, stored procedures, triggers — including relationships across repository boundaries (service calls, APIs, shared tables).
- Serves that graph over MCP, the open protocol spoken by Claude, GitHub Copilot and other AI agents. Agents stop guessing — they query.
- Answers graph questions in minutes instead of days, with source attribution, using a fraction of the tokens an agent needs without a graph.
- Hooks into CI/CD: every merge and release updates the graph automatically.

Example: "What depends on the column customers.status?" returns the methods and stored procedures that write it, the procedures that read it, the triggers it fires, and the services in other repositories that call those — as a path through the graph, every edge with its source.

Typical questions komjo answers:

- What breaks if we change this? (impact analysis / blast radius)
- Which code uses this database table? (database lineage)
- How are these two parts of the system connected? (call paths)
- Where is the dead code?
- Give our AI agents real context about the codebase.

## Supported stacks

- Analyzers per language, for example C# / .NET, TypeScript, SQL and database schemas. Every further language is an analyzer plugin, not a rewrite; many more to come.
- Applicants can register interest for C / C++, Go, Java / JVM, Python, Rust and others, and for SQL Server, PostgreSQL, MySQL / MariaDB and Oracle.

## Trust and deployment

- Enterprise option runs entirely on the customer's infrastructure — on-prem or private cloud. Source code does not leave the customer's systems. A managed cloud (SaaS) option is also offered.
- One isolated knowledge base per repository and snapshot version, in its own tenant. Versions (main, releases, PR branches) sit side by side in the graph and can be compared; edges such as service calls, shared tables and API calls connect repositories.
- Every fact carries its source. Access to the graph is read-only: komjo answers questions and never changes code.

## Early adopter program

For teams with a large or complex codebase. Limited spots; cohorts are composed by stack. Participants get:

- A pilot on one of their own repositories with an isolated, per-tenant knowledge base.
- An MCP endpoint their team and AI agents can query from day one.
- A direct line to the people building komjo, and real influence on what ships next.

What komjo expects: a real repository and a real question, honest feedback in a short call every few weeks, and the patience of an early adopter. If the pilot succeeds, being named as a reference or telling the story together, always by agreement. Conditions are agreed per team; early adoption is a partnership, not a sales channel.

Process: apply (five minutes, form on the start page), intro call to match stack and question against the analyzers, then the pilot on the team's own repository.

## Links

- [Apply for the early-adopter program](https://komjo.ai/#apply)
- [Launch news and newsletter sign-up](https://komjo.ai/#updates)
- [Pixiray](https://pixiray.ai/)
- [Imprint](https://komjo.ai/impressum/)
- [Privacy policy](https://komjo.ai/datenschutz/): no cookies, no tracking; forms and mailing list run with Microsoft as processor in EU data centres; launch news and newsletter use double opt-in and can be cancelled any time.
- Contact: hello@pixiray.ai (a WhatsApp chat link is on the start page)