Overview

Fireworks Nexus brings frontier open models and model routers to coding harnesses, agents, and gateways, with Fireworks usage visibility and per-user spend caps for supported serverless models.

New to Nexus? Start with the Quickstart. For the product overview, see the Nexus product page.

Quickstart

Connect your first harness, app, or gateway.

FireConnect

One command to route a coding harness through Fireworks.

FireRouter

One model ID that picks a model for each turn.

How Nexus fits together

Connection and model choice are independent. Bring a coding tool, application, or LLM gateway. Then call a Fireworks model directly, or let FireRouter choose one for each new user turn.

1. Connect

   FireConnect for coding harnesses, APIs and SDKs for your own code, or LLM Gateways for LiteLLM, Portkey, and others. Prefer not to install FireConnect? Every harness has manual setup.

2. Choose models

   Select Open Models directly, or let FireRouter pick one for each user turn.

3. Add closed models

   FireRouter calls Claude and GPT models on your own Anthropic, OpenAI, or Amazon Bedrock account. Connect Provider Keys once where available, or send the key with each request.

4. Operate

   Usage and Cost shows Claude Code session estimates and dashboard usage for Fireworks models. Spend Limits caps supported Fireworks serverless usage.