Cohere Models via OpenRouter
Ever Works does not ship a standalone Cohere plugin. Cohere models, including the Command R family, are available through the OpenRouter plugin. This page describes how to configure and use Cohere models for work generation, content creation, and AI conversations.
Related source files:
| File | Purpose |
|---|---|
packages/plugins/openrouter/src/openrouter.plugin.ts | OpenRouter AI provider plugin |
packages/plugin/src/ai/reasoning.utils.ts | Reasoning configuration utilities |
What is Cohere?
Cohere is an enterprise AI platform that offers models optimized for business applications. Their Command R models are designed for retrieval-augmented generation (RAG), multilingual content, and structured data extraction -- all capabilities that align well with work generation workflows.
Available Cohere Models
Cohere models are available on OpenRouter with the cohere/ prefix:
| Model ID (OpenRouter) | Description | Strengths |
|---|---|---|
cohere/command-r-plus | Command R+ | Highest quality, RAG-optimized |
cohere/command-r | Command R | Balanced quality and cost |
cohere/command | Command | Cost-effective for simpler tasks |
cohere/command-light | Command Light | Fastest, lowest cost |
Model availability on OpenRouter changes over time. Check openrouter.ai/models for the current list of available Cohere models and pricing.
Why Consider Cohere for Works?
Cohere models have specific strengths relevant to work generation:
| Strength | Benefit for Ever Works |
|---|---|
| RAG optimization | Excels at generating content from source material |
| Multilingual support | Strong performance across many languages |
| Structured output | Reliable JSON generation for item data |
| Grounding | Can cite sources when generating content |
| Enterprise focus | Consistent, reliable output quality |
Configuration
Setting Up Cohere Models
- Navigate to Settings > Plugins in the Ever Works dashboard.
- Ensure the OpenRouter plugin is enabled (enabled by default).
- Enter your OpenRouter API key.
- Set model fields to Cohere model IDs:
| Setting | Recommended Value |
|---|---|
| Default Model | cohere/command-r |
| Simple Tasks Model | cohere/command-light |
| Standard Tasks Model | cohere/command-r |
| Complex Tasks Model | cohere/command-r-plus |
Environment Variables
PLUGIN_OPENROUTER_API_KEY=sk-or-...
PLUGIN_OPENROUTER_DEFAULT_MODEL=cohere/command-r
PLUGIN_OPENROUTER_SIMPLE_MODEL=cohere/command-light
PLUGIN_OPENROUTER_MEDIUM_MODEL=cohere/command-r
PLUGIN_OPENROUTER_COMPLEX_MODEL=cohere/command-r-plus
Architecture
Cohere models follow the same access path as all other OpenRouter models:
The OpenRouter plugin uses an OpenAI-compatible API endpoint. OpenRouter translates requests to the Cohere-native format internally, so no special handling is needed in the plugin.
Tiered Model Strategy
Cohere's model lineup maps naturally to the Ever Works tier system:
Mixed-Provider Configuration
Combine Cohere with other providers for optimal cost-quality balance:
| Tier | Model | Provider | Rationale |
|---|---|---|---|
| Simple | openai/gpt-5-nano | OpenAI | Fastest for tags |
| Standard | cohere/command-r | Cohere | Strong at structured content |
| Complex | cohere/command-r-plus | Cohere | RAG-optimized for source-based generation |
Capabilities
| Capability | Command R+ | Command R | Command Light |
|---|---|---|---|
| Structured output | Yes | Yes | Yes |
| Streaming | Yes | Yes | Yes |
| Tool calling | Yes | Yes | Limited |
| Vision | No | No | No |
| Embeddings | No (via OpenRouter) | No (via OpenRouter) | No |
| Multilingual | Strong | Strong | Good |
Embedding Limitation
Cohere offers embedding models (e.g., embed-english-v3.0), but these are not currently accessible through the OpenRouter plugin in Ever Works. If you need embeddings:
- Use the OpenAI plugin (
text-embedding-3-small). - Use the Google Gemini plugin (built-in embedding support).
- Use the Ollama plugin with a local embedding model.
Comparison with Other Providers
| Aspect | Cohere (via OpenRouter) | OpenAI (direct) | Anthropic (direct) |
|---|---|---|---|
| Setup | OpenRouter key | OpenAI key | Anthropic key |
| RAG optimization | Strong | Good | Good |
| Multilingual | Very strong | Good | Good |
| Structured output | Reliable | Very reliable | Reliable |
| Vision | No | Yes (GPT-4o+) | Yes (Claude) |
| Embeddings | Not via OpenRouter | Yes | No |
| Cost | Moderate | Varies by model | Varies by model |
Use Cases for Cohere in Works
Multilingual Works
Cohere's multilingual capabilities make it well-suited for works that serve content in multiple languages. Command R models handle cross-language content generation without significant quality degradation.
Source-Heavy Works
When works are built from extensive source material (web scraping, PDFs, Notion pages), Cohere's RAG optimization helps the model ground its output in the provided sources rather than hallucinating information.
Structured Data Extraction
Cohere models reliably produce structured JSON output, which is critical for generating work item data (names, descriptions, categories, tags, metadata).
Troubleshooting
| Issue | Cause | Solution |
|---|---|---|
| Model not found | Model ID incorrect or model discontinued | Verify at openrouter.ai/models |
| Authentication error | Invalid OpenRouter API key | Regenerate key at openrouter.ai |
| Rate limit exceeded | Too many requests | Check OpenRouter plan limits |
| Poor multilingual output | Wrong model selected | Use Command R or Command R+ for multilingual |
| Embedding request fails | Cohere embeddings not available via OpenRouter | Use OpenAI or Ollama for embeddings |
| JSON parsing errors | Structured output not enforced | Ensure pipeline uses structured output mode |
Further Reading
- OpenRouter Plugin -- complete OpenRouter configuration
- AI Provider Plugins -- all available AI providers
- DeepSeek Models -- another provider accessible via OpenRouter
- xAI Grok Models -- Grok models via OpenRouter
- Plugin Settings -- settings resolution and configuration modes