| name | doc |
| description | Ingest an external markdown reference doc into memhub so it is RAG-searchable in chunks. The first ingest in a repo auto-enables doc chunks in default recall (decision 90). Trigger on: "ingest this doc", "add this spec to memhub", "make this doc searchable", "remember this reference doc", "/doc".
|
| framework | memhub |
| framework_version | 1.0.0 |
| last_updated | "2026-05-18T00:00:00.000Z" |
Ingest a local markdown file into this repo's .memhub/project.sqlite
as an external reference document, chunked by heading and embedded so
it is retrievable through the same SQL+RAG hybrid recall as facts,
decisions, and tasks. The first successful ingest in a repo
auto-enables [retrieval] include_docs_in_default (decision 90), so
afterward a plain recall call can surface a doc chunk directly — gated
by the [retrieval.scoring] doc_min_rerank_score relevance floor.
This is the Codex counterpart to the Claude Code /doc skill. Both
call into the same memhub doc CLI and memhub.doc_add MCP tool;
they differ only in the agent identifier on write-side telemetry.
Use this when the user wants to pull pieces of a long doc across a
session instead of loading the whole thing into context.
Preconditions
.memhub/ exists in the working repo (run /check-init if unsure).memhub binary on PATH.- The file is a local markdown file the user pointed at.
If a precondition is missing, surface it and stop.
Invocation
Prefer the memhub.doc_add MCP tool when available:
memhub.doc_add(file="<absolute or repo-relative path>.md")
Optional title="<override>" — defaults to first heading or file name.
CLI fallback:
memhub doc add "<path>.md" --json
memhub doc ls --json
memhub doc show <id|path> --json
memhub doc rm <id|path> --json
Add --global to ls, show, or rm to manage docs in the
machine-global store instead of this repo's (mirrors doc add --global; requires memhub global enable in this repo — see
/global). Without --global these only touch the repo store; global
doc ids are per-global-DB, independent of repo ids:
memhub doc ls --global --json
memhub doc show <id|path> --global --json
memhub doc rm <id|path> --global --json
Re-ingest semantics
doc add on an already-ingested path: unchanged content (same
SHA-256) is a no-op (status: "unchanged"); changed content replaces
every chunk and refreshes embeddings/FTS (status: "updated"). Safe to
re-run after the user edits the file.
Retrieving from the doc
Plain recall already surfaces ingested doc chunks once this repo's
first doc add has run (decision 90). To restrict a query to docs
alone, scope explicitly:
memhub.recall(query="<question>", source_types=["doc"])
Plain recall also returns an available_docs count — doc chunks that
did NOT surface this call; when it is non-zero and the question is
design/spec/architecture-flavored, a doc-scoped follow-up can still be
worth running (judgment, not reflex). Doc hits return with
source_type: "doc_chunk" and a <title> — <section breadcrumb>
title.
Scope boundary
memhub is not a general knowledge base. Ingested docs are per-repo,
user-pointed reference material scoped to this repo's work. Ingest
what the user asks for, not arbitrary files.
Notes
doc_add is a direct write (no review queue); logged to
writes_log.- Docs are excluded from
memhub export (disk-backed, re-ingestable
cache) — re-run doc add on another machine.
- Embeddings populate only in
hybrid mode; fts mode ingests
chunks + FTS, vector recall starts after memhub index rebuild.
