herolib/examples/ai/openai

OpenRouter Examples - Proof of Concept

Overview

This folder contains example scripts demonstrating how to use the OpenAI client (herolib.clients.openai) configured to work with OpenRouter.

• Goal: Show how to send messages to OpenRouter models using the OpenAI client, run a two-model pipeline for code enhancement, and illustrate multi-model usage.
• Key Insight: The OpenAI client is OpenRouter-compatible by design - simply configure it with OpenRouter's base URL (https://openrouter.ai/api/v1) and API key.

---

Configuration

All examples configure the OpenAI client to use OpenRouter by setting:

• URL: https://openrouter.ai/api/v1
• API Key: Read from OPENROUTER_API_KEY environment variable
• Model: OpenRouter model IDs (e.g., qwen/qwen-2.5-coder-32b-instruct)

Example configuration:

playcmds.run(
    heroscript: '
        !!openai.configure
            name: "default"
            url: "https://openrouter.ai/api/v1"
            model_default: "qwen/qwen-2.5-coder-32b-instruct"
    '
)!

---

Example Scripts

1. openai_init.vsh

• Purpose: Basic initialization example showing OpenAI client configured for OpenRouter.
• Demonstrates: Client configuration and simple chat completion.
• Usage:

examples/ai/openai/openai_init.vsh

---

2. openai_hello.vsh

• Purpose: Simple hello message to OpenRouter.
• Demonstrates: Sending a single message using client.chat_completion.
• Usage:

examples/ai/openai/openai_hello.vsh

• Expected output: A friendly "hello" response from the AI and token usage.

---

3. openai_example.vsh

• Purpose: Demonstrates basic conversation features.
• Demonstrates:
  • Sending a single message
  • Using system + user messages for conversation context
  • Printing token usage
• Usage:

examples/ai/openai/openai_example.vsh

• Expected output: Responses from the AI for both simple and system-prompt conversations.

---

4. openai_two_model_pipeline.vsh

• Purpose: Two-model code enhancement pipeline (proof of concept).
• Demonstrates:
  • Model A (Qwen3 Coder) suggests code improvements.
  • Model B (morph-v3-fast) applies the suggested edits.
  • Tracks tokens and shows before/after code.
  • Using two separate OpenAI client instances with different models
• Usage:

examples/ai/openai/openai_two_model_pipeline.vsh

• Expected output:
  • Original code
  • Suggested edits
  • Final updated code
  • Token usage summary

---

Environment Variables

Set your OpenRouter API key before running the examples:

export OPENROUTER_API_KEY="sk-or-v1-..."

The OpenAI client automatically detects when the URL contains "openrouter" and will use the OPENROUTER_API_KEY environment variable.

---

Notes

1. No separate OpenRouter client needed - The OpenAI client is fully compatible with OpenRouter's API.
2. All scripts configure the OpenAI client with OpenRouter's base URL.
3. The two-model pipeline uses two separate client instances (one per model) to demonstrate multi-model workflows.
4. Scripts can be run individually using the v -enable-globals run command.
5. The two-model pipeline is a proof of concept; the flow can later be extended to multiple files or OpenRPC specs.