ai-messages-to-prompt Edge Function — Ai

ai-messages-to-prompt — Convert messages to a formatted LLM prompt string

Serialises a structured {role, content} message array into a raw prompt string for open-source LLMs (ChatML, Llama 2, Alpaca, or plain text formats).

---

API

POST /api/ai-messages-to-prompt

Request body

Field	Type	Required	Default	Description
messages	{role: "system"│"user"│"assistant", content: string}[]	✅	—	Conversation messages
format	"chatml" | "llama2" | "alpaca" | "plain"	❌	"chatml"	Prompt template format
addGenerationPrompt	boolean	❌	true	Append the model's generation prefix

Success response (200)

{
  "success": true,
  "data": {
    "prompt": "<|im_start|>system\nYou are helpful.<|im_end|>\n<|im_start|>user\nHello<|im_end|>\n<|im_start|>assistant\n",
    "format": "chatml",
    "messageCount": 2
  }
}

Error responses

Code	HTTP	When
INVALID_INPUT	400	Invalid role or missing messages
INTERNAL_ERROR	500	Unknown format

---

Usage

cURL

curl -X POST "$FUNCTION_URL" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "system", "content": "You are helpful."},
      {"role": "user", "content": "What is 2+2?"}
    ],
    "format": "llama2"
  }'

TypeScript / JavaScript (HTTP)

const response = await fetch(FUNCTION_URL, {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ messages, format: 'chatml' }),
});
const { data } = await response.json();
const rawPrompt = data.prompt;
// Send rawPrompt to your local LLM inference engine

Direct import (Node / Bun / Deno)

import { aiMessagesToPrompt } from '@aerostack/functions/ai-messages-to-prompt';

const { prompt } = aiMessagesToPrompt({ messages, format: 'alpaca' });

---

Format Examples

ChatML

<|im_start|>system
You are helpful.<|im_end|>
<|im_start|>user
Hello<|im_end|>
<|im_start|>assistant

Llama 2

[INST] <<SYS>>
You are helpful.
<</SYS>>

Hello [/INST]

Alpaca

### Instruction:
You are helpful.

### Input:
Hello

### Response:

Plain

System: You are helpful.
User: Hello
Assistant:

---

Use Cases

• Formatting a conversation history for a local Llama/Mistral/Phi model that requires a specific prompt template.
• Building a prompt serialisation layer in an agent framework that supports multiple open-source LLMs.
• Generating training data in the correct format for fine-tuning open-source models.

---

Notes

• Only system, user, and assistant roles are accepted — other role values throw an error.
• addGenerationPrompt: false omits the trailing generation prefix (useful for training data generation).
• For Llama 2 multi-turn: system is injected into the first [INST] block only.