Your knowledge,
down to the marrow.

Marrow is a quiet, self-hosted knowledge base — a restore guarantee you can audit, an export bundle you can read by hand, and no vendor lock-in.

The unshowy things
that compound.

Block editor, by the book

Slash commands, wiki-links, tables, callouts, code. Nothing you haven't seen — everything done well.

Backlinks that mean something

Every page knows what links to it — the backlink index is rebuilt on every save.

Every save versioned

Append-only history: every save is a new revision you can roll back to.

Cmd+K, and that's it

Search is the navigation. Page names, bodies, and properties — one field, one answer, anywhere in a workspace.

Your data, yours to take

Pages and history live in Postgres, attachments on disk or S3/R2 — and any workspace exports to a readable Markdown + JSON zip.

Comment, don't co-write

Threads, not cursors. For the 95% of docs that aren't meant to be live-edited.

Postmortem — ingest saturation, Nov 4

At 02:14 UTC we saw queue depth climb past 40k. Retries stacked, the downstream writer started dropping, and alerts fired 90 seconds later.

# Postmortem — ingest saturation, Nov 4

At 02:14 UTC we saw queue depth climb past 40k.
Retries stacked, the downstream writer started
dropping, and alerts fired 90 seconds later.

> [!action] Add backpressure shedding on the
> writer before GA.

```shell
$ kubectl top pods -n ingest
writer-0  cpu=980m  mem=1.9Gi
```

Related:: [[RFC-0142]] [[Runbook / Ingest]]

One compose file.
Done.

One Docker Compose file brings up three services: the API image, the web image, and a Postgres database. Pages and history live in Postgres; attachments on the filesystem or S3/R2. Back it up by backing up the database and attachment store.

# 1. Pull the API image
$ docker pull ghcr.io/marrow-software/marrow-api:latest

# 2. Configure and bring up the stack (db + api + web)
$ cp .env.prod.example .env   # set POSTGRES_PASSWORD, SECRET_KEY, …
$ docker compose -f docker-compose.prod.yml up -d

✓ api on :8000  ✓ web on :3000