EKOS

•
28
•

Compiles enterprise source code, schemas, and documentation into an append-only, evidence-backed knowledge graph for AI agents via MCP.

EKOS operates as an open-source enterprise knowledge compiler that turns source code, SQL, Git history, and technical documentation into a deterministic, queryable semantic model for AI agents. By observing systems without directly interpreting them, it compiles disparate organizational data into an append-only ledger where every fact is grounded in verifiable evidence. AI coding assistants like Claude Code interface with this reconstructed knowledge over a read-only Model Context Protocol (MCP) server, avoiding direct interaction with sensitive production systems while slashing token consumption and boosting text-to-SQL accuracy.

주요 기능

0128 GitHub stars
02Append-only semantic ledger linking every fact, relationship, and event to immutable, verifiable evidence
03Read-only Model Context Protocol (MCP) interface isolating AI agents from raw enterprise infrastructure
04Built-in secret and PII redaction that automatically prevents tokens, credentials, and sensitive files from being indexed
05Deterministic knowledge compiler that converts source code, SQL, and documentation into a Canonical Knowledge Model
06Extensive AST and catalog parsing for languages like Python, Rust, Java, and SQL alongside databases like ClickHouse and Oracle

사용 사례

01Improving text-to-SQL generation accuracy by providing models with compiled schema knowledge and relationship graphs
02Equipping AI coding agents with deep repository context and call graphs while drastically cutting token consumption
03Preserving institutional knowledge and analyzing legacy codebases during enterprise architectural migrations