What is EKOS?
EKOS is a compiler for enterprise knowledge: it turns source code, databases, documents and binaries into a traceable knowledge layer AI agents can query.
EKOS compiles enterprise systems into structured, evidence-backed knowledge that AI agents query through the Model Context Protocol (MCP).
It observes the systems you already have — source code, Git history, SQL schemas and migration scripts, the live catalogs of Oracle, PostgreSQL and ClickHouse databases, GitHub issues, Confluence, PDFs, Pentaho jobs and, in a separately licensed extension build, compiled .NET and JVM binaries — and runs them through deterministic compiler passes. The result is a Canonical Knowledge Model stored in an append-only ledger. Every object, relationship and conclusion in that ledger points back to the evidence it was derived from.
What you get
- A knowledge layer, not a search index. Tables, services, functions, pipelines, people and documents become typed objects with typed relationships (
DependsOn,Calls,ForeignKey,Extends, …). - Answers with receipts. Every conclusion carries evidence and provenance: file, line span, the analyzer that produced it, and the pipeline run that wrote it.
- History. The ledger is append-only, so you can ask what was true at any past time (
ekos_statewithat,ekos diff, EKLAS OF). - Cheap repeated questions. Compile once, query many times. On a 2,022-file open-source repository, answering real questions from the ledger used 67–93% fewer tokens than grep-based search over the source.
- Disagreements kept, decisions made by people. When two sources disagree, both claims are kept for a person to settle. Uncertain matches, conflicts and the source-model sign-off are human-only decisions that no MCP tool can make. See Human Review.
- No cloud dependency. The compiler, ledger and MCP server are one Rust binary; the optional web console (Python API, React UI) runs beside it, on one machine or in a cloud. The compiler passes need no LLM; LLM use is opt-in and limited to specific, labelled steps, and every pipeline run logs how many model calls it made.
Where to go next
- New here? Read The Problem, then run the Quickstart.
- Want the vocabulary? Core Concepts.
- Want the internals? System Overview.