glm-vision

Pi extension that gives non-vision GLM models (z.ai) image understanding by routing images through a GLM vision model.

What this is

glm-vision intercepts Pi read tool results that include images when you are using the zai provider. It sends those images to a GLM vision model (default: glm-4.6v) and returns a combined text description to the active text-only GLM model.

Features

• Automatic image interception for z.ai GLM text models
• Ordered multi-image support with per-image + combined summaries
• Prompt presets for OCR, UI, diagrams, and code, plus custom prompts
• Response cache keyed by image hash, prompt, and model
• Safe fallback behavior that preserves original images on error

Install

Install the published npm package with Pi:

pi install npm:glm-vision

Pin a specific version when you want reproducible installs:

pi install npm:[email protected]

Install into the current project instead of your user Pi settings:

pi install npm:glm-vision -l

Or install from GitHub:

pi install git:github.com/eiei114/glm-vision

Try it without permanently installing:

pi -e npm:glm-vision

Quick start

After installing, start a Pi session (or run locally with pi -e .) and confirm the extension loaded:

/glm-vision:status

Then ask Pi to read an image:

Read ./screenshots/checkout-error.png and explain what is wrong with this UI.

Usage notes (summary)

• Vision models: uses the zai provider and defaults to glm-4.6v. See the full list and availability checks in docs/usage.md.
• Multiple images: images are sent in their original order and referenced as Image 1, Image 2, and so on.
• Limits: maxImages defaults to 4. If more images are present, the first maxImages are described and the remainder are skipped.
• Fallback behavior: if no image data is available or the vision request fails, glm-vision returns an error message while preserving the original image blocks so Pi can continue its fallback path.

Command examples:

| Command | Description | | --- | --- | | /glm-vision:status | Show status, model, prompt mode, and cache stats. | | /glm-vision:on | Enable image description. | | /glm-vision:off | Disable image description. | | /glm-vision:model | Open a TUI picker to switch vision models. | | /glm-vision:mode | Open a TUI picker to switch prompt presets. | | /glm-vision:cache-status | Show cache status. |

Legacy space forms such as /glm-vision on and /glm-vision glm-4.6v remain available for compatibility. More details, including presets, configuration, and troubleshooting, live in docs/usage.md.

Package contents

| Path | Purpose | | --- | --- | | src/ | Pi extension entrypoint (src/index.ts) | | docs/ | Usage, examples, release, and maintainer docs | | scripts/ | Upstream model watcher utilities | | tests/ | Vitest coverage for core behavior | | .github/workflows/ | CI, publish, auto-release, upstream watch |

Development

npm install
npm run lint
npm run typecheck
npm test
npm run validate:package

Optional upstream model checks:

npm run check:upstream

Release

This package uses npm Trusted Publishing (OIDC) via GitHub Actions.

npm version patch
git push origin HEAD

See docs/release.md and RELEASE.md for the full maintainer checklist.

Maintainer docs

• docs/examples.md — usage examples
• docs/template-checklist.md — Pi extension template alignment checklist

Security

Pi packages can execute code with your local permissions. Review extensions before installing third-party packages.

For vulnerability reporting, see SECURITY.md.

Links

• npm: https://www.npmjs.com/package/glm-vision
• GitHub: https://github.com/eiei114/glm-vision
• Issues: https://github.com/eiei114/glm-vision/issues

License

MIT