Fatrocu Server

Fatrocu Server provides a lightweight HTTP service for processing invoice images using the local llama-cli model. It listens on port 1842 by default and exposes a single endpoint:

• POST /process – multipart/form-data with fields image, model, temp, n_predict, ctx_size.

The server forwards the request to the llama-cli executable, runs the specified model and returns the raw JSON output.

Building

cd C:/Users/PC/Desktop/fatrocu-server
cargo build --release

The binary is placed at target/release/fatrocu-server.exe.

Running

# Default port 1842
target\release\fatrocu-server.exe

# Custom port
FATROCU_SERVER_PORT=8080 target\release\fatrocu-server.exe

The service binds to 0.0.0.0 and logs the listening address.

API

POST /process

The request must be multipart/form-data containing:

• image – the image file to process.
• model (optional) – model identifier, defaults to the built‑in model.
• temp (optional) – sampling temperature, default 0.2.
• n_predict (optional) – maximum tokens, default 8.
• ctx_size (optional) – context window size, default 256.

The server constructs a llama-cli command similar to:

llama-cli \
  --model <model>.gguf \
  --mmproj <model>-mmproj-f16.gguf \
  --image <temp file> \
  --temp <temp> \
  --threads 4 \
  --gpu-layers 0 \
  --n-predict <n_predict> \
  --ctx-size <ctx_size>

The command’s stdout is returned as the HTTP response with Content-Type: application/json. Errors are returned as HTTP 500 with the error message.

Configuration

• FATROCU_SERVER_PORT – TCP port to listen on (default 1842).
• Ensure llama-cli.exe is available in the system PATH or adjust the source code to point to its location.

License

MIT License – see the LICENSE file.