Sources
Sources are the systems and artifacts EKOS observes; each has a connector.
A source is observed by a connector (an Observer) that returns content-addressable artifacts and never interprets them.
| Category | Sources |
|---|---|
| Code | Rust, Python (incl. SQLAlchemy models and PySpark chains), Java (incl. the iDempiere/ADempiere application model and SQL built in method bodies), JavaScript/TypeScript, Elixir, Perl |
| SQL files | PostgreSQL, MySQL, SQL Server, Snowflake, Databricks, ClickHouse, Trino and Oracle dialects: tables, views, PL/pgSQL and PL/SQL routines and triggers; dated migration scripts as schema history |
| Live databases | Oracle, PostgreSQL and ClickHouse catalogs, declared in [[databases]] and read once per build in a read-only session (see below) |
| Pipelines | dbt projects; Pentaho .ktr/.kjb |
| Documents | PDF, DOCX, text, Markdown (front matter kept as properties), HTML, email; Confluence |
| History | Git commits and the ticket ids they name; GitHub issues and pull requests |
| Structure | Cargo.toml crate topology; GitHub Actions workflows; dependency manifests |
| Binaries | Compiled .NET and JVM, in a separately licensed extension build (see Indexing .NET Applications) |
| Scaffolded (mock-tested only) | Salesforce, SAP, Microsoft Fabric, Snowflake |
A project names its inputs as [[sources]] (a repository or directory each); objects record the source they came from, which is what a per-source read scope filters on; an object with no source (a rollup, say) is hidden from a source-restricted reader.
Live databases
A [[databases]] entry holds a connection, a read-only user and the name of the environment variable that holds the password, never the password itself. ekos build reads the catalog — tables, columns, keys, check constraints, views, routines, triggers — and reconciles it with the database's own object census; ekos database test checks the connection and compares that census with what was compiled. A write probe must be refused, and a user that could write is reported. For Oracle, an optional profile adds optimizer statistics (p0), exact row counts (p1) or one aggregate scan per table (p2: null counts, number and date bounds, text lengths); a text value is never read out. PostgreSQL and ClickHouse use statistics only.
Secrets and PII are redacted before anything is stored — see Security Model. Details per source: Knowledge Sources.