Skip to main content

Fireworks AI Models via OpenRouter

Ever Works does not ship a standalone Fireworks AI plugin. Fireworks AI is one of the inference providers that OpenRouter routes requests to when you select open-source models. This page explains how Fireworks AI fits into the Ever Works AI pipeline and how to take advantage of its fast inference.

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 Fireworks AI?

Fireworks AI is a fast-inference platform for open-source AI models. It specializes in optimized serving of models like Llama, Mistral, and Qwen, often achieving faster response times than other providers. When you use open-source models through OpenRouter in Ever Works, Fireworks AI may be one of the backend providers handling your requests.

How OpenRouter Uses Fireworks AI

OpenRouter acts as a routing layer. When you select a model such as meta-llama/llama-3.3-70b-instruct, OpenRouter can route the request to whichever provider offers the best combination of price, speed, and availability -- including Fireworks AI.

You do not choose Fireworks AI directly. Instead, you choose a model, and OpenRouter handles provider selection automatically.

Provider Preferences

OpenRouter allows users to set provider preferences on their account. If you want to ensure requests are routed to Fireworks AI specifically:

  1. Log in to your OpenRouter account.
  2. Navigate to account settings.
  3. Set provider preferences for your preferred models.

These preferences are applied at the OpenRouter level, outside of Ever Works configuration.

Models Commonly Served by Fireworks AI

Fireworks AI typically serves open-source models. Common models that may be routed to Fireworks AI include:

Model ID (OpenRouter)Description
meta-llama/llama-3.3-70b-instructMeta Llama 3.3 70B
meta-llama/llama-4-scoutMeta Llama 4 Scout
mistralai/mistral-largeMistral Large
qwen/qwen-2.5-72b-instructQwen 2.5 72B

Configuration in Ever Works

Since Fireworks AI is accessed transparently through OpenRouter, no special configuration is needed beyond setting up the OpenRouter plugin:

  1. Enable the OpenRouter plugin (enabled by default).
  2. Enter your OpenRouter API key.
  3. Select the open-source models you want to use.

Environment Variables

PLUGIN_OPENROUTER_API_KEY=sk-or-...
PLUGIN_OPENROUTER_DEFAULT_MODEL=meta-llama/llama-3.3-70b-instruct

Performance Characteristics

Fireworks AI is known for fast inference times. This benefits Ever Works in several ways:

BenefitDescription
Faster work generationItems are processed more quickly during pipeline execution
Lower latency AI chatConversational responses arrive faster
Better throughputMore items can be processed in parallel

Speed Comparison

ProviderRelative SpeedCost
Fireworks AI (via OpenRouter)FastPer-token, varies
Groq (direct)Very fast (custom hardware)Free tier + paid
Ollama (local)Depends on hardwareFree
OpenAI (direct)StandardPer-token

If raw inference speed is your top priority and you want open-source models, consider the Groq plugin as a direct alternative. Groq uses custom LPU hardware and offers the fastest inference for supported models.

Tiered Strategy with Fireworks-Served Models

TierModelWhy
Simplemeta-llama/llama-3.3-70b-instruct:freeZero cost for tags and classifications
Standardmeta-llama/llama-3.3-70b-instructGood quality, fast via Fireworks AI
Complexmeta-llama/llama-4-maverickBest open-source quality for full pages

Capabilities

Models served by Fireworks AI through OpenRouter support standard capabilities:

CapabilitySupported
Structured output (JSON)Yes
StreamingYes
Tool callingModel-dependent
VisionModel-dependent
EmbeddingsGenerally no

Comparison: Fireworks AI vs Groq vs Ollama

AspectFireworks AI (via OpenRouter)Groq (direct plugin)Ollama (direct plugin)
Access methodTransparent via OpenRouterDedicated pluginDedicated plugin
API keyOpenRouter keyGroq keyNone required
SpeedFastVery fastHardware-dependent
CostPer-tokenFree tier availableFree
PrivacyCloud-hostedCloud-hostedLocal
Model controlOpenRouter selects providerFixed to GroqFull local control
EmbeddingsNoNoYes

Troubleshooting

IssueCauseSolution
Cannot select Fireworks directlyOpenRouter handles routingUse OpenRouter provider preferences for routing control
Slow response timesNot routed to Fireworks AICheck OpenRouter provider preferences
Rate limit errorsProvider-level rate limitsUpgrade your OpenRouter plan or switch providers
Model quality variesDifferent providers may serve slightly different quantizationsPin provider preferences in your OpenRouter account

Further Reading