How to set prompt for llama-server

[Qmeimei10086]

I found argument "-p, --prompt PROMPT prompt to start generation with; for system message, use -sys" for llama-cli command, but i don't find something like for llama-server command,could someone tell me how to set prompt for llama-server command !!!!!

[Riya-877]

llama-server does not support a -p/--prompt flag like llama-cli.
Instead, prompts are passed through HTTP requests after the server is running.

How to use prompts with llama-server

1. Start the server:

./llama-server -m model.gguf

2. Send a prompt via API request:

Completion endpoint:

curl http://localhost:8080/completion \
  -H "Content-Type: application/json" \
  -d '{
        "prompt": "Explain AI in simple terms",
        "n_predict": 100
      }'

Chat-style endpoint (recommended):

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
        "messages": [
          {"role": "system", "content": "You are a helpful assistant"},
          {"role": "user", "content": "Explain AI in simple terms"}
        ]
      }'

Key difference

• llama-cli: uses -p and -sys flags
• llama-server: uses JSON fields (prompt or messages) in API requests

So in llama-server, you don’t pass prompts as command-line arguments — you send them in the request body.