Skip to main content

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:

FilePurpose
packages/plugins/openrouter/src/openrouter.plugin.tsOpenRouter AI provider plugin
packages/plugin/src/ai/reasoning.utils.tsReasoning 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)DescriptionStrengths
cohere/command-r-plusCommand R+Highest quality, RAG-optimized
cohere/command-rCommand RBalanced quality and cost
cohere/commandCommandCost-effective for simpler tasks
cohere/command-lightCommand LightFastest, lowest cost
note

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:

StrengthBenefit for Ever Works
RAG optimizationExcels at generating content from source material
Multilingual supportStrong performance across many languages
Structured outputReliable JSON generation for item data
GroundingCan cite sources when generating content
Enterprise focusConsistent, reliable output quality

Configuration

Setting Up Cohere Models

  1. Navigate to Settings > Plugins in the Ever Works dashboard.
  2. Ensure the OpenRouter plugin is enabled (enabled by default).
  3. Enter your OpenRouter API key.
  4. Set model fields to Cohere model IDs:
SettingRecommended Value
Default Modelcohere/command-r
Simple Tasks Modelcohere/command-light
Standard Tasks Modelcohere/command-r
Complex Tasks Modelcohere/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:

TierModelProviderRationale
Simpleopenai/gpt-5-nanoOpenAIFastest for tags
Standardcohere/command-rCohereStrong at structured content
Complexcohere/command-r-plusCohereRAG-optimized for source-based generation

Capabilities

CapabilityCommand R+Command RCommand Light
Structured outputYesYesYes
StreamingYesYesYes
Tool callingYesYesLimited
VisionNoNoNo
EmbeddingsNo (via OpenRouter)No (via OpenRouter)No
MultilingualStrongStrongGood

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

AspectCohere (via OpenRouter)OpenAI (direct)Anthropic (direct)
SetupOpenRouter keyOpenAI keyAnthropic key
RAG optimizationStrongGoodGood
MultilingualVery strongGoodGood
Structured outputReliableVery reliableReliable
VisionNoYes (GPT-4o+)Yes (Claude)
EmbeddingsNot via OpenRouterYesNo
CostModerateVaries by modelVaries 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

IssueCauseSolution
Model not foundModel ID incorrect or model discontinuedVerify at openrouter.ai/models
Authentication errorInvalid OpenRouter API keyRegenerate key at openrouter.ai
Rate limit exceededToo many requestsCheck OpenRouter plan limits
Poor multilingual outputWrong model selectedUse Command R or Command R+ for multilingual
Embedding request failsCohere embeddings not available via OpenRouterUse OpenAI or Ollama for embeddings
JSON parsing errorsStructured output not enforcedEnsure pipeline uses structured output mode

Further Reading