Skip to main content

Together AI Models via OpenRouter

Ever Works does not include a standalone Together AI plugin. Together AI models -- including open-source models like Llama, Mistral, and Qwen -- are accessed through the OpenRouter plugin. This page explains how to configure and use Together AI models for work generation.

Related source files:

FilePurpose
packages/plugins/openrouter/src/openrouter.plugin.tsOpenRouter AI provider plugin
packages/plugin/src/ai/reasoning.utils.tsReasoning config for open-source models
apps/internal-cli/src/commands/config/ai-providers/ai-provider-registry.service.tsProvider registry with model lists

What is Together AI?

Together AI is a cloud platform for running open-source AI models. It provides fast inference for popular models such as Meta Llama, Mistral, Qwen, and others. Through OpenRouter, these models become available in Ever Works without any additional plugin.

Available Models

Together AI models are available on OpenRouter with the together/ or provider-specific prefixes:

Model ID (OpenRouter)Base ModelDescription
meta-llama/llama-3.3-70b-instructLlama 3.3 70BHigh-quality open-source model
meta-llama/llama-4-scoutLlama 4 ScoutLatest Llama generation
meta-llama/llama-4-maverickLlama 4 MaverickLarge-context Llama 4 variant
mistralai/mistral-largeMistral LargeStrong multilingual model
qwen/qwen-2.5-72b-instructQwen 2.5 72BCompetitive open-source model
note

Together AI hosts many open-source models. OpenRouter may route to Together AI or another provider based on availability and cost. Check openrouter.ai/models for current availability.

Configuration

Setting Up Together AI Models

  1. Navigate to Settings > Plugins in the Ever Works dashboard.
  2. Ensure the OpenRouter plugin is enabled.
  3. Enter your OpenRouter API key.
  4. Set model fields to open-source model IDs:
SettingRecommended Value
Default Modelmeta-llama/llama-3.3-70b-instruct
Simple Tasks Modelmeta-llama/llama-3.3-70b-instruct:free
Standard Tasks Modelmeta-llama/llama-3.3-70b-instruct
Complex Tasks Modelmeta-llama/llama-4-maverick

Environment Variables

PLUGIN_OPENROUTER_API_KEY=sk-or-...
PLUGIN_OPENROUTER_DEFAULT_MODEL=meta-llama/llama-3.3-70b-instruct
PLUGIN_OPENROUTER_SIMPLE_MODEL=meta-llama/llama-3.3-70b-instruct:free
PLUGIN_OPENROUTER_COMPLEX_MODEL=meta-llama/llama-4-maverick

Free Models

OpenRouter offers some models with a :free suffix (e.g., meta-llama/llama-3.3-70b-instruct:free). These use donated compute and have rate limits, but cost nothing. They are suitable for simple pipeline tasks like tag generation and short descriptions.

Architecture

OpenRouter automatically selects the best provider for each model request. When you specify a model like meta-llama/llama-3.3-70b-instruct, OpenRouter may route to Together AI, Fireworks AI, or another provider hosting that model, depending on availability and cost.

Tiered Model Strategy

Open-source models from Together AI work well in cost-optimized configurations:

TierUse CaseRecommended ModelCost
SimpleTags, classificationsmeta-llama/llama-3.3-70b-instruct:freeFree
StandardSummaries, listingsmeta-llama/llama-3.3-70b-instructLow
ComplexFull page generationmeta-llama/llama-4-maverickModerate

Mixing with Commercial Models

You can combine open-source models for cost-sensitive tasks with commercial models for quality-critical tasks:

TierModelProviderCost Level
Simplemeta-llama/llama-3.3-70b-instruct:freeMeta (via Together AI)Free
Standardopenai/gpt-4oOpenAIModerate
Complexanthropic/claude-sonnet-4AnthropicHigher

Capabilities

CapabilityLlama 3.3 70BLlama 4 Scout/Maverick
Structured outputYesYes
StreamingYesYes
Tool callingYesYes
VisionNoModel-dependent
EmbeddingsNoNo

Embedding Limitation

Open-source models on Together AI generally do not provide embedding endpoints through OpenRouter. If your workflow requires embeddings for semantic search:

  • Use the OpenAI plugin for embeddings alongside Together AI models for generation.
  • Or use the Ollama plugin with a local embedding model.
  • Or use the Google Gemini plugin which includes embedding support.

Alternative: Local Models via Ollama

If you prefer to run open-source models locally instead of through Together AI, the Ollama plugin provides direct access:

ApproachTogether AI (via OpenRouter)Ollama (Local)
HostingCloud (Together AI servers)Self-hosted
API keyOpenRouter API key requiredNo API key needed
CostPer-token pricing (some free)Free (your hardware)
SpeedFast (datacenter GPUs)Depends on local hardware
PrivacyData sent to cloudData stays local
Available modelsWide selectionAny Ollama-compatible model

Troubleshooting

IssueCauseSolution
Model not availableModel removed or renamed on OpenRouterCheck current model list at openrouter.ai
Rate limited on free modelFree tier has strict rate limitsSwitch to paid model or wait
Poor output qualityModel not suited for taskTry a larger model or commercial alternative
Timeout on large requestsOpen-source models may be slower for long outputsReduce maxTokens or use a faster model
JSON parsing errorsSome models have weaker structured outputEnable structured output mode or use a model with better JSON support

Further Reading