Rabun is a CLI. It does not ship a workspace. Declare pipelines in TOML, write wiki trees to git, and serve MCP on stdio or 127.0.0.1:8080. Burton is the optional screen.

YAML pipeline

rabun.toml is the overlay. The [etl] table names the warehouse work dir and pipelines. [[apps]] starts sibling writers.

[etl]
work_dir = ".rabun/etl"

[[etl.pipelines]]
name = "contacts"
publication = "contacts"
mode = "cdc"

[[apps]]
name = "wiki"
command = "rabun-wiki"
args = ["drain"]

rabun etl run
rabun etl ingest
rabun-wiki drain
rabun serve

Extractors drop JSONL under .rabun/etl. rabun etl ingest loads it. See ETL and companions.

Wiki without a UI

rabun-wiki is the git writer: Markdown+YAML into objects/, inbox/, docs/, and synthesis/. MCP tools wiki_list, wiki_read, and wiki_search read RABUN_WIKI_ROOT (alias BURTON_WIKI_ROOT, default data/wiki). You never open a browser for that loop.

For a vault that never talks to Burton at all, use curator.

What still needs a database

Workers that claim Burton rows (SMS, mail, feeds, forms) still want Postgres. Headless means no session UI, not “no data.” Warehouse ETL can run against any publisher that owns pg_publication.