Skip to content

container-images.md

Latest commit

 

History

History
125 lines (85 loc) · 5.04 KB

container-images.md

File metadata and controls

125 lines (85 loc) · 5.04 KB

Container images (GitHub Container Registry)

Official TalkPipe images are built in CI and published to GitHub Container Registry (GHCR).

Use Docker or Podman the same way; the examples below use docker—substitute podman if you prefer.

Set an image tag once (pin a release version or use latest):

IMAGE=ghcr.io/sandialabs/talkpipe:latest
# IMAGE=ghcr.io/sandialabs/talkpipe:0.11.7   # example pinned version

Multi-platform releases

When a GitHub release is published, CI pushes a multi-platform manifest:

  • linux/amd64 — typical x86_64 servers, Intel Macs (Linux VM), Windows (Docker Desktop / WSL2)
  • linux/arm64 — Apple Silicon Macs, many ARM64 Linux servers (for example AWS Graviton)

Your client pulls the correct architecture automatically. To list platforms in a release image (replace the tag with the version you use):

docker manifest inspect ghcr.io/sandialabs/talkpipe:0.11.7 | grep architecture

Non-release workflow runs (pushes and pull requests) still publish images for CI, but those builds are single-platform (linux/amd64 only) for faster pipelines. For reproducible, portable deployments, use an image built from a release tag.

Tags and versions

Images are tagged from the same metadata CI uses for releases and branches. Useful patterns:

Tag When to use
latest Latest stable release (not pre-releases tagged as alpha/beta). Good for quick starts; pin a version for production.
experimental Latest pre-release (alpha/beta) GitHub release, when applicable.
0.11.7 (example) Exact release (full semantic version). Best for reproducible deploys.
0.11 Major.minor line—floats to the newest patch in that minor series for that release stream.
Branch or PR refs Built from CI on pushes/PRs; amd64 only; useful for testing unreleased commits.
Git SHA Immutable pointer to the exact image built for a commit.

Replace 0.11.7 with the version you want from the releases page.

Pull an image

Public images can be pulled without logging in to GHCR (subject to GitHub’s normal rate limits for anonymous pulls):

docker pull ghcr.io/sandialabs/talkpipe:0.11.7
docker pull ghcr.io/sandialabs/talkpipe:latest

If your environment requires authentication (for example private registry mirrors or organization policy):

docker login ghcr.io
# Use a GitHub personal access token with `read:packages` where prompted for a password.

Example commands

The image includes TalkPipe’s published CLIs. Override the container command after the image name.

HTTP servers default to loopback inside the container, so published ports (-p) from the host will not work unless the process binds to all interfaces. Use --host 0.0.0.0 (chatterlang_workbench) or -o 0.0.0.0 (chatterlang_serve, serverag). Then open http://127.0.0.1:<port> on the host (if you map the port to 127.0.0.1 only, prefer that URL over http://localhost:..., which may resolve to IPv6 and miss the forward).

Help and non-network CLIs

docker run --rm "$IMAGE" chatterlang_workbench --help
docker run --rm "$IMAGE" chatterlang_serve --help
docker run --rm "$IMAGE" serverag --help
docker run --rm "$IMAGE" makevectordatabase --help
docker run --rm "$IMAGE" chatterlang_script --help
docker run --rm "$IMAGE" chatterlang_reference_generator --help
docker run --rm "$IMAGE" talkpipe_plugins --help

ChatterLang workbench (web UI, default port 4143)

docker run --rm -p 127.0.0.1:4143:4143 "$IMAGE" \
  chatterlang_workbench --host 0.0.0.0

chatterlang_serve (default port 2025)

Add --script, --form-config, and other options as needed; binding uses -o for host:

docker run --rm -p 127.0.0.1:2025:2025 "$IMAGE" \
  chatterlang_serve -o 0.0.0.0 -p 2025

serverag — RAG web UI (default port 2026)

Mount your LanceDB directory and pass --path (see also makevectordatabase and serverag):

docker run --rm -p 127.0.0.1:2026:2026 -v "$PWD/mydb:/app/data:ro" "$IMAGE" \
  serverag --path /app/data -o 0.0.0.0

makevectordatabase (one-shot; no --host)

docker run --rm -v "$PWD/docs:/app/docs" -v "$PWD/mydb:/app/data" "$IMAGE" \
  makevectordatabase "/app/docs/*.md" --path /app/data

RAG: end-to-end (build DB, then serve)

docker run --rm -v "$PWD/docs:/app/docs" -v "$PWD/mydb:/app/data" "$IMAGE" \
  makevectordatabase "/app/docs/*.md" --path /app/data

docker run --rm -p 127.0.0.1:2026:2026 -v "$PWD/mydb:/app/data" "$IMAGE" \
  serverag --path /app/data -o 0.0.0.0

Build details (maintainers)

CI configuration and tag behavior are summarized in .github/CICD.md.