[LIVE] [PaloAltoNetworks/panw-api-ollama] 1 endpoint(s)
ChatGPT
API Proxy/ChatGPT
2,133 characters
Status: [LIVE]
Endpoints found:
https://pan.dev/prisma-airs/scan/api/
Source: https://github.com/PaloAltoNetworks/panw-api-ollama
# panw-api-ollama

A proxy that sits in front of [Ollama](https://ollama.com/) and sends prompts and responses to the Palo Alto Networks Prisma AIRS AI Runtime API (API Intercept) for scanning before they reach the model or the user.
> **Community proof-of-value project.** panw-api-ollama is an open-source sample integration, not a supported Palo Alto Networks product. Use it to evaluate and demo AI Runtime scanning with Ollama. Report issues here on GitHub, not through Palo Alto Networks support.
## What it does
Point your client (for example [OpenWebUI](https://openwebui.com/)) at the proxy instead of Ollama. The proxy forwards everything to Ollama, and on the routes below it scans the content first. Depending on your AI Runtime security profile, a scan can block or mask:
- Prompt injection
- Data leakage (DLP, PII)
- Malicious code and URLs
- Harmful or toxic content
Blocked prompts never reach the model. Blocked or masked responses are replaced before they reach the client, including in streaming mode.
## What is scanned
Only these routes are scanned:
| Route | Scanned |
|---|---|
| `POST /api/chat` | prompt (all messages) and response |
| `POST /api/generate` | prompt, `system`, `template`, and response |
| `POST /api/embeddings`, `POST /api/embed` | input |
Every other path is forwarded to Ollama **without scanning**: model management (`/api/tags`, `/api/pull`, ...), and the OpenAI/Anthropic compatibility endpoints (`/v1/chat/completions`, `/v1/messages`, `/v1/embeddings`). If clients must not reach those, set `PASSTHROUGH_DISABLED=1` and every unhandled path returns `404`.
## Install
### Docker (recommended)
See the [Docker setup guide](docker/README.md).
### From source
1. Build:
```bash
git clone https://github.com/PaloAltoNetworks/panw-api-ollama.git
cd panw-api-ollama
cargo build --release
```
2. Get an AI Runtime API Intercept key and