BorrowedCompute

BorrowedCompute is a retro-styled client for chatting with Ollama models, generating images from community InvokeAI servers, and for discovering new public instances via Shodan. A launcher provides a unified entry point on Linux, macOS, and Windows.

Installation

1. Clone the repository

   git clone <REPO_URL>
   cd BorrowedCompute

2. Install Python dependencies

   cd scripts
   python -m pip install -r requirements.txt

   On Windows, use the Python launcher if needed:

   py -m pip install -r requirements.txt

3. Prepare data files

   • Copy data/ollama.endpoints.example.csv to data/ollama.endpoints.csv and add any known Ollama servers.
   • Copy data/invoke.endpoints.example.csv to data/invoke.endpoints.csv for InvokeAI hosts.
   • (Optional) Create data/config.json to store your Shodan API key and Discord bot settings (DISCORD_BOT_TOKEN, BORROWEDCOMPUTE_WORKING_DIR). A template is available at data/config.example.json.

Usage

Launcher

Use the launcher for a graphical menu with matrix rain effects.

• Linux/macOS: ./launcher.sh
• Windows: launcher.bat

The launcher now groups actions by API. Pick LLM Chat or Image Generation, then choose a provider-specific task such as launching BorrowedCompute or running a Shodan scan scoped to that API. Endpoint refreshes still update the corresponding CSV files under data/.

The Configure menu is organized by capability: Configure Chat contains Ollama server/model selection, while Configure Image Generation contains InvokeAI and Automatic1111 server/model selection.

Press the highlighted number to choose an option. The launcher restores the console on exit.

Chat Client

Run the chat client directly if desired:

python3 scripts/BorrowedCompute.py

The client loads servers from data/ollama.endpoints.csv (for chat) and data/invoke.endpoints.csv (for image generation), pings them to sort by latency, and stores conversations under data/conversations. Logs produced with the /print command are written to data/logs.

Image Generation

InvokeAI and Automatic1111 image generation are available from the main BorrowedCompute menu. The client discovers API-compatible servers, lists the models each host exposes, and walks through prompt configuration (including width/height, steps, CFG scale, scheduler or sampler, and optional seed). Generated images are downloaded to data/images/ along with per-render metadata.

BorrowedCompute renders image previews directly in the terminal using chafa.py (with Pillow) and will fall back to the chafa CLI if the Python bindings are unavailable.

Discord Bot (optional)

An optional Discord bot mirrors the launcher menus with per-user, ephemeral interactions. Both chat and image-generation flows are available in Discord through the same menu groupings used by the launcher.

1. Install dependencies if you haven't already:

   cd scripts
   pip3 install -r requirements.txt

2. Add your Discord settings to data/config.json:

   {
     "DISCORD_BOT_TOKEN": "<your token>",
     "BORROWEDCOMPUTE_WORKING_DIR": "optional/path/if/scripts/aren't/in/PWD"
   }

3. Start the bot:

   python3 discord_bot.py

The bot exposes a /config command with a compact, private configuration menu. Choose Shodan, Chat, or Imagine, then select from servers and models that pass live compatibility checks. Send prompts with /chat (Ollama) or /imagine (InvokeAI or Automatic1111). Shodan scans run directly from Discord.

Server and model selections are shared through data/menu_state.json. Configure a server and model in the terminal app, exit it, and the Discord bot will use those selections after validating availability. Discord selections are written back to the same file; data/discord_preferences.json remains a compatibility fallback for older saved preferences.

Shodan Scan

Scan for new supported servers and verify existing ones:

python3 scripts/shodanscan.py [--debug] [--limit N] [--existing-limit N]

The script requires a Shodan API key. Results are appended to data/ollama.endpoints.csv and data/invoke.endpoints.csv, enriched with metadata such as hostnames, organisation, ISP, location, and (for InvokeAI) available models. Existing entries are checked for availability and ping time.

Development

Run a basic syntax check on all Python scripts:

python3 -m py_compile scripts/*.py

Data Layout

• data/ollama.endpoints.csv – list of known Ollama endpoints.
• data/invoke.endpoints.csv – list of known InvokeAI endpoints.
• data/images/ – InvokeAI image outputs and metadata files.
• data/conversations/ – per-model conversation history.
• data/logs/ – saved transcripts.
• data/config.json – optional configuration (SHODAN_API_KEY).
• data/discord_preferences.json – persisted Discord server and model choices.

License

This project is provided as-is without any warranty. Use it responsibly.