n8n-nodes-openrouter-cache

OpenRouter Chat Model node for n8n with prompt caching support. Drop-in replacement for the built-in OpenRouter Chat Model that injects cache_control markers to reduce costs up to 90% on Anthropic and Gemini models.

Features

• Prompt caching for Anthropic and Gemini models via OpenRouter (system message + optional last user message)
• Configurable TTL: 5-minute default or 1-hour extended cache
• Cache breakpoints: System message only, or system + last user message
• Caching enabled by default — the whole point of this node
• Tool call fix preserved from upstream (fixes empty arguments from Anthropic-via-OpenRouter)
• Uses existing openRouterApi credential — no extra setup

Install

In your n8n instance:

1. Go to Settings → Community Nodes
2. Click Install
3. Enter n8n-nodes-openrouter-cache-chat-model
4. Click Install

Usage

1. Add the OpenRouter Cache Chat Model node to your workflow
2. Connect it to an AI Agent or AI Chain node
3. Select your model (e.g., anthropic/claude-sonnet-4-20250514 or google/gemini-2.5-flash)
4. Caching is enabled by default — configure under Options if needed

Options

| Option | Default | Description | |--------|---------|-------------| | Enable Prompt Caching | true | Toggle caching on/off | | Cache TTL | 5 Minutes | 5 Minutes or 1 Hour (1h has 2x write cost on Anthropic) | | Cache Breakpoints | System Message Only | System Message Only or System + Last User Message |

How it works

The node wraps the standard ChatOpenAI from LangChain with a custom fetch interceptor that:

1. Intercepts outgoing API requests
2. Transforms system message content from string to content blocks with cache_control: { type: "ephemeral" }
3. Optionally marks the last user message for caching too
4. Sends the modified request to OpenRouter

On the response side, it also fixes empty tool call arguments (a known issue with Anthropic models via OpenRouter).

Which models benefit?

| Provider | Caching | Notes | |----------|---------|-------| | Anthropic (Claude) | Explicit — this node adds it | Min 1024-4096 tokens depending on model | | OpenAI (GPT) | Automatic | No modification needed, but node won't break anything | | DeepSeek | Automatic | Same as OpenAI | | Google (Gemini 2.5) | Explicit — this node adds it | Min 1,028 tokens (Flash) / 2,048 tokens (Pro). Fixed ~3-5min TTL |

Verifying cache hits

Check the OpenRouter Activity dashboard after running your workflow twice:

• First run: cache_write_tokens > 0 (cache established)
• Second run: cached_tokens > 0 (cache hit, reduced cost)

Development

npm install
npm run build

For local testing, copy dist/ to your n8n custom nodes directory:

cp -r dist/ ~/.n8n/custom/node_modules/n8n-nodes-openrouter-cache-chat-model/dist/
cp package.json ~/.n8n/custom/node_modules/n8n-nodes-openrouter-cache-chat-model/

License

MIT