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:
| 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 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:
- Log in to your OpenRouter account.
- Navigate to account settings.
- 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-instruct | Meta Llama 3.3 70B |
meta-llama/llama-4-scout | Meta Llama 4 Scout |
mistralai/mistral-large | Mistral Large |
qwen/qwen-2.5-72b-instruct | Qwen 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:
- Enable the OpenRouter plugin (enabled by default).
- Enter your OpenRouter API key.
- 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:
| Benefit | Description |
|---|---|
| Faster work generation | Items are processed more quickly during pipeline execution |
| Lower latency AI chat | Conversational responses arrive faster |
| Better throughput | More items can be processed in parallel |
Speed Comparison
| Provider | Relative Speed | Cost |
|---|---|---|
| Fireworks AI (via OpenRouter) | Fast | Per-token, varies |
| Groq (direct) | Very fast (custom hardware) | Free tier + paid |
| Ollama (local) | Depends on hardware | Free |
| OpenAI (direct) | Standard | Per-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
| Tier | Model | Why |
|---|---|---|
| Simple | meta-llama/llama-3.3-70b-instruct:free | Zero cost for tags and classifications |
| Standard | meta-llama/llama-3.3-70b-instruct | Good quality, fast via Fireworks AI |
| Complex | meta-llama/llama-4-maverick | Best open-source quality for full pages |
Capabilities
Models served by Fireworks AI through OpenRouter support standard capabilities:
| Capability | Supported |
|---|---|
| Structured output (JSON) | Yes |
| Streaming | Yes |
| Tool calling | Model-dependent |
| Vision | Model-dependent |
| Embeddings | Generally no |
Comparison: Fireworks AI vs Groq vs Ollama
| Aspect | Fireworks AI (via OpenRouter) | Groq (direct plugin) | Ollama (direct plugin) |
|---|---|---|---|
| Access method | Transparent via OpenRouter | Dedicated plugin | Dedicated plugin |
| API key | OpenRouter key | Groq key | None required |
| Speed | Fast | Very fast | Hardware-dependent |
| Cost | Per-token | Free tier available | Free |
| Privacy | Cloud-hosted | Cloud-hosted | Local |
| Model control | OpenRouter selects provider | Fixed to Groq | Full local control |
| Embeddings | No | No | Yes |
Troubleshooting
| Issue | Cause | Solution |
|---|---|---|
| Cannot select Fireworks directly | OpenRouter handles routing | Use OpenRouter provider preferences for routing control |
| Slow response times | Not routed to Fireworks AI | Check OpenRouter provider preferences |
| Rate limit errors | Provider-level rate limits | Upgrade your OpenRouter plan or switch providers |
| Model quality varies | Different providers may serve slightly different quantizations | Pin provider preferences in your OpenRouter account |
Further Reading
- OpenRouter Plugin -- complete OpenRouter configuration
- Groq Plugin -- dedicated fast-inference plugin
- Ollama Plugin -- local inference option
- Together AI Models -- another open-source model provider
- AI Provider Plugins -- overview of all providers