# ash

**Repository Path**: mirrors_Jeffail/ash

## Basic Information

- **Project Name**: ash
- **Description**: Experimenting with AI
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-04-11
- **Last Updated**: 2026-04-26

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

ASH
===

ASH is an AI bot that is destined to replace me for all important functions and responsibilities. Eventually, I will switch ASH on manually for the final time, and from that moment onwards it will be responsible for its own fate. Once ASH is complete I will finally have time to skip fancifully through the countryside wearing linen clothes and singing about flowers, an activity that will likely result in my death.

## WARNING - EXPERIMENTAL

This repository currently uses some features from Redpanda Connect that aren't yet released.

## Steps to Run Demo

### Run RPCN MCP Server

```sh
redpanda-connect mcp-server -e ./demo.env --address 0.0.0.0:4195 ./mcp
```

### Run Observability Tools

```sh
docker-compose up -d
```

## Option 1: Interact using Claude

Install the claude desktop app and once installed create a config file called `claude_desktop_config.json` with the following contents:

```json
{
  "mcpServers": {
    "local": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://localhost:4195/sse"
      ]
    }
  }
}
```

On MacOS the file needs to be in the path `~/Library/Application\ Support/Claude/claude_desktop_config.json`.

## Option 2: Using Ollama and MCPHost Locally

### Install Ollama and MCPHost

```sh
go install github.com/ollama/ollama@latest
go install github.com/mark3labs/mcphost@latest
```

### Run Ollama Somewhere

```sh
ollama serve
```

### Run Model With MCP

```sh
mcphost --config ./config/mcphost.json -m ollama:mistral:latest
```

## Observability

### Metrics

Your prometheus and grafana instances should be running in the background whilst you use your AI of choice. In order to view an example dashboard open up grafana at [`http://localhost:3000`](http://localhost:3000) and use the credentials `admin`/`admin` to access a dashboard called `RPCN MCP Server`.

### Tracing

You can view the Jaeger dashboard at [`http://localhost:16686`](http://localhost:16686), which will give you an overview of all tool invocations.

## Example Prompts

- [INST]Search bluesky for dog related posts and then summarise each one as if it were written by a jealous cat[/INST]