Skip to content

docs: the Context Hub ships with the Pipecat CLI - #1055

Open
markbackman wants to merge 1 commit into
mainfrom
docs/context-hub-ships-with-cli
Open

docs: the Context Hub ships with the Pipecat CLI#1055
markbackman wants to merge 1 commit into
mainfrom
docs/context-hub-ships-with-cli

Conversation

@markbackman

Copy link
Copy Markdown
Contributor

The Context Hub now ships inside pipecat-ai[cli], so uv tool install "pipecat-ai[cli]" is the whole install. These pages still taught a --with co-install, weighed two install paths against each other, and warned that --with replaces the tool environment — a choice that no longer exists.

Depends on pipecat-ai/pipecat#5122; land after that releases.

What changed

Page
api-reference/context-hub.mdx Setup is the CLI install. Removed the --with warning and the whole "Which install to choose" section; kept the uvx path for using the hub without the CLI
api-reference/cli/context-hub.mdx ## Installation down to a line. Added install's exit codes and that re-running converges rather than skipping
api-reference/cli/overview.mdx context-hub moves from "(requires pipecat-ai-context-hub)" to "(built in)"; the --with example reverts to pipecatcloud only
api-reference/cli/init.mdx New "Context Hub setup" section, plus the --context-hub/--no-context-hub flag
pipecat/get-started/build-your-next-bot.mdx "Add the Context Hub" becomes "The Context Hub" — init does it for you
pipecat/get-started/quickstart.mdx Drops the co-install; says why quickstart doesn't stop to ask

Two additions beyond the rename

pipecat init performs the setup, and no page said so. It registers the MCP server with the coding agent CLIs it finds and offers to build the index. That is the largest user-visible change in the release, so init.mdx gains a section and the getting-started pages stop instructing people to do it by hand.

install's exit codes are documented, including 3. It is the only way a caller distinguishes "configured a client" from "printed the config for you to paste" — both print, and neither is an error. Cursor, VS Code, and Zed always take the second path.

Corrected figures

The old text said "~2 min first run" in one place and "allow several minutes" in another, and no page mentioned the disk cost.

Measured cold: 17,359 records, 922 MB, in 2m46s and 3m48s across two runs on the same machine — both with a warm model cache, so a genuine first run also pays a ~186 MB model download. A later refresh with one source changed took 77s.

So the first build is stated as a floor — upwards of three minutes and about 900 MB of disk — rather than a target. The total is dominated by embedding ~17k records on the CPU, so it tracks single-core speed; a point estimate would be wrong for most readers. This matches the 3+ minutes the CLI prompt shows. Refresh is documented separately at one to two minutes, since it re-ingests only sources whose contents moved.

Also corrected

context-hub.mdx claimed a standalone install plus the CLI's copy costs "a second copy on disk — the two environments share very little." They share the index, which is the ~900 MB part; only the package duplicates.

Checks

mint isn't installed locally, so internal links were verified by hand — every target exists. Worth a mint broken-links run before merge.

@github-actions

github-actions Bot commented Aug 4, 2026

Copy link
Copy Markdown
Contributor

🔍 Mintlify preview for this branch: https://daily-docs-context-hub-ships-with-cli.mintlify.site

`uv tool install "pipecat-ai[cli]"` provides `pipecat context-hub`, so setup is
the CLI install. The pages that taught a `--with` co-install, and the section
comparing the two installs, describe a choice that no longer exists.

`pipecat init` now performs the setup on its coding-agent path — registering the
MCP server and offering to build the index — which no page covered. `install`'s
exit codes are documented alongside it, since `3` is how a caller tells "printed
the config to paste" from "configured a client".

First-build cost is stated as a floor, upwards of three minutes and about 900 MB
of disk: it varies with single-core speed, and a first run on a cold machine also
downloads the embedding models. Refresh is given separately, since it re-ingests
only the sources that moved.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant