overthinkos/notebook-openrouter

notebook-openrouter -- OpenRouter API integration data candy

Candy Properties

| Property | Value | |----------|-------| | Dependencies | (none) | | Packages | (none) | | Services | (none) | | Volumes | workspace -> /workspace (from jupyter) | | Data | data/openrouter -> workspace volume, dest: openrouter | | env_require | OPENROUTER_API_KEY — API key for OpenRouter LLM inference | | Install files | (none) |

How It Works

This is a data candy with an env_require declaration — it uses the data: field to map notebooks to a volume, and env_require: to declare that the OPENROUTER_API_KEY environment variable must be present:

info: "OpenRouter API integration notebook collection"

env_require:
  - name: OPENROUTER_API_KEY
    description: "API key for OpenRouter LLM inference"

data:
  - src: data/openrouter
    volume: workspace
    dest: openrouter

At build time, the contents of data/openrouter/ are staged into /data/workspace/openrouter/ inside the box.

At deploy time, charly config copies the staged data into the workspace volume at <workspace>/openrouter/. The dest: openrouter field places the notebooks in a subdirectory rather than the volume root.

The env_require declaration is stored as an OCI label (ai.opencharly.env_require). At charly config time, if OPENROUTER_API_KEY is not present in the resolved environment, a warning is printed:

Warning: jupyter-ml-notebook requires OPENROUTER_API_KEY (API key for OpenRouter LLM inference) — not set

This is the first candy in the project to use the env_require feature.

Included Notebooks

3 Jupyter notebooks covering OpenRouter API usage with the free model qwen/qwen3.6-plus:free:

| Notebook | Topic | Features | |----------|-------|----------| | 00_OpenRouter_Basics.ipynb | API Basics | Authentication, simple chat completion, multi-turn conversation, error handling, usage/cost tracking | | 01_OpenRouter_Models.ipynb | Model Selection | List all models, filter free models, inspect model details (pricing, context, modality), compare responses across models | | 02_OpenRouter_Inference.ipynb | Practical Inference | Structured JSON output, text summarization, code generation (with exec), reasoning with thinking tokens, translation pipeline |

Manifest: notebooks.yaml — structured catalog with title, description, and ordering.

Free Tier Rate Limit Gotcha

The free model qwen/qwen3.6-plus:free has strict rate limits (~1 request per minute). All notebooks include an openrouter_chat() / chat() retry helper that handles 429 responses with exponential backoff:

def openrouter_chat(model, messages, retries=5, delay=15):
    for attempt in range(retries):
        response = requests.post(...)
        if response.status_code == 429:
            wait = delay * (attempt + 1)
            print(f"Rate limited, waiting {wait}s...")
            time.sleep(wait)
            continue
        response.raise_for_status()
        return response.json()

The retry helper also handles transient 502 upstream errors from Alibaba (the Qwen provider) where the response is 200 but missing the choices key.

Providing the API Key

The OPENROUTER_API_KEY must be injected into the container environment:

# Via charly config (persisted in quadlet)
charly config jupyter-ml-notebook -e OPENROUTER_API_KEY=sk-or-v1-...

# Via workspace .env file
echo "OPENROUTER_API_KEY=sk-or-v1-..." >> ~/project/.env

# Via direnv .secrets (GPG-encrypted)
charly secrets gpg set OPENROUTER_API_KEY sk-or-v1-...

Usage

# charly.yml
jupyter-ml-notebook:
  candy:
    - jupyter-ml
    - notebook-templates
    - notebook-finetuning
    - notebook-ollama
    - notebook-llm-on-supercomputers
    - notebook-openrouter
    # ... other candies

# Deploy with API key
charly config jupyter-ml-notebook -e OPENROUTER_API_KEY=sk-or-v1-...
charly start jupyter-ml-notebook
# Open http://localhost:8888 -> navigate to openrouter/

Used In Boxes

• /charly-jupyter:jupyter-ml-notebook

Related Skills

• /charly-image:layer — data field documentation, env_require/env_accept field documentation
• /charly-core:charly-config — data provisioning and env_require warning during charly config setup
• /charly-core:deploy — volume backing configuration
• /charly-jupyter:notebook-ollama — sibling data candy pattern (Ollama integration notebooks)
• /charly-jupyter:notebook-finetuning — sibling data candy pattern (Unsloth fine-tuning notebooks)
• /charly-jupyter:notebook-templates — sibling data candy pattern (starter notebooks)
• /charly-jupyter:notebook-llm-on-supercomputers — sibling data candy pattern (TU Wien course notebooks)
• /charly-jupyter:jupyter-ml-notebook — the box that includes this candy

When to Use This Skill

Use when the user asks about:

• The notebook-openrouter candy or its contents
• OpenRouter API tutorial notebooks or integration examples
• How to provide OPENROUTER_API_KEY to Jupyter containers
• The env_require feature (this is the first real consumer)
• Free tier rate limits on qwen/qwen3.6-plus:free
• Structured output, reasoning with thinking tokens, or model comparison via OpenRouter

Related

• /charly-eval:eval — declarative testing (eval: block, charly eval box, charly eval live)