README.md

---
title: GLM Visual Variable Runtime
emoji: 👁️
colorFrom: indigo
colorTo: blue
sdk: docker
app_port: 8000
short_description: OpenAI-compatible visual runtime gateway
---

# GLM Visual Variable Runtime

`glm-5.2-visual-runtime` is a training-free multimodal gateway that exposes an OpenAI-compatible API while using GLM-5.2 as the reasoning engine and lazy visual lenses for images, OCR, palettes, objects, charts, and presentations.

The MVP in this repository is contract-first:

- FastAPI endpoints for `/v1/responses`, `/v1/chat/completions`, `/v1/files`, `/health`, and `/ready`.
- Persistent visual assets with hash-based deduplication and lazy cached views.
- Deterministic `test` mode providers, plus replaceable interfaces for GLM-5.2, GLM-5V-Turbo, and GLM-OCR.
- CPU-only ingest fingerprinting with Pillow, OpenCV, and NumPy.
- A TypeScript PptxGenJS renderer service for editable PowerPoint artifacts.
- No training, no fine-tuning, and no GLM-5.2 self-hosting.

## Quick Start

```bash
cp .env.example .env
docker compose up --build
```

API docs are available at `http://localhost:8000/docs`.

## Local Tests

```bash
cd apps/api
python3 -m venv .venv
. .venv/bin/activate
pip install -e ".[test]"
pytest ../../tests
```

## Example

```python
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="test")
response = client.responses.create(
    model="glm-5.2-visual-runtime",
    input=[{
        "role": "user",
        "content": [{"type": "input_text", "text": "Hello"}],
    }],
)
print(response.output_text)
```

## Hugging Face Space

The deployment target is a Docker Space. Set the following secrets in the Space for cloud mode:

- `GLM_API_KEY`
- `VISION_API_KEY`
- `OCR_API_KEY`
- object store and database credentials

Without secrets, set `VISUAL_RUNTIME_MODE=test` to use deterministic fake providers.