# Run Hello ShinrAI locally

Hello ShinrAI is the open-source local playground for first requests, protected LLM chat and transparent API debugging. It binds to your computer and requires no database, frontend build or local model.

## Launch

Install [uv](https://docs.astral.sh/uv/getting-started/installation/) once, then run the reviewed release:

```bash
uvx --from https://github.com/Innovius-ai/hello-shinrai/releases/download/v0.1.4/hello_shinrai-0.1.4-py3-none-any.whl hello-shinrai
```

The browser opens at `http://127.0.0.1:8765`. To use another port or a headless machine:

```bash
uvx --from https://github.com/Innovius-ai/hello-shinrai/releases/download/v0.1.4/hello_shinrai-0.1.4-py3-none-any.whl hello-shinrai --port 8877 --no-browser
```

This guide is pinned to the tested [v0.1.4 release](https://github.com/Innovius-ai/hello-shinrai/releases/tag/v0.1.4). Change both version values in the wheel URL when upgrading. The [repository setup guide](https://github.com/Innovius-ai/hello-shinrai) also includes conventional Python and Docker instructions.

## Connect ShinrAI

Open **Connections**, enter your ShinrAI application key and confirm the base URL. The managed URL defaults to `https://api.getshinrai.com`; sandbox and on-premises URLs are editable.

Connecting reads `/v2/capabilities` and `/v2/usage`. The app shows the key's current balance, models and Standard, Batch and Real-time tier access. Real-time is labelled **fast**. A failed connection is reported separately from a valid key without Real-time entitlement.

The ShinrAI key enables text protection, documents and scans, Azure-shaped requests to ShinrAI, AWS Comprehend compatibility and Google DLP compatibility.

## Add protected chat

Chat additionally needs an LLM provider key. Choose Innovius, OpenAI, Kie or Custom, then confirm its model-list URL, Chat Completions or Responses URL and model ID. A model listed in a catalogue is not automatically verified for Responses or vision input.

Every turn protects the complete local conversation before calling the LLM. The debugger shows the original local input, ShinrAI result, exact protected LLM payload, provider reply and locally restored reply. A protection failure stops the request. Streaming, cancellation and split-token restoration are supported.

TXT, PDF, DOCX, PNG, JPEG and BMP attachments can supply protected text. Vision-capable models can also receive page images rendered exclusively from ShinrAI's protected PDF. Visual protection covers detected text; it does not hide faces, barcodes or non-text identifiers.

## Debug and export

The trace records request identifiers, findings, mappings, warnings, returned usage, local response time, server timing when available, first-token time and document cleanup. It never records credentials.

**Export safe** removes original content and replacement maps. **Export with private data** keeps them after an explicit warning. Everything else remains in memory and disappears when the process stops. Optional secure saving uses the operating system credential store and falls back to memory-only operation when no supported store is available.

The API Explorer permits only reviewed processing operations and safe processing paths discovered from the deployment's OpenAPI document. It does not expose Dashboard, account, OAuth, MCP or operator routes and does not automatically retry billable requests.