Skip to main content
Start with Coding Harnesses for setup by client. This page covers explicit model changes, closed-model credentials, and Foundry support.

​
Add an explicit model

Connecting a harness already registers coding-ready Fireworks models, including auto. Use --model only when you want an explicit model or router:
Replace claude and glm-latest with the harness and model you want. On Claude Code, --model adds the model to /model without replacing native Anthropic tiers. After changing a model, follow the restart or reopen instructions in Coding Harnesses. For model IDs, aliases, fast tiers, pinned versions, image support, and US-only models, see Open Models.

​
Connect credentials for closed model families

Your Fireworks key pays for the open-model leg. A closed model family also needs a credential for its provider.
For a route containing Claude or another Anthropic model, connect an account-level Anthropic key in Provider Keys when possible. Developers then need only their Fireworks API key, which fireconnect login configures for them. The table below shows harness-local alternatives and current FireConnect constraints.
FireConnect currently refuses routes that require Anthropic credentials on Cursor IDE, Copilot App, Copilot CLI, and DeepSeek Harness. For Cursor IDE, this is a client header limitation. For Copilot and DeepSeek, it is a FireConnect integration gap. OpenAI models use an account-level OpenAI Provider Key. FireConnect does not have a local OpenAI-key flag. For direct HTTP calls, see APIs and SDKs. To control the balance between predicted quality and cost, see Routing Preferences.

​
Foundry compatibility

On the Foundry path, pass the Azure deployment name, such as FW-GLM-5.2, instead of a short Fireworks ID. Model routers are not available on the Foundry path. Claude Code, DeepSeek Harness, Copilot App, and Copilot CLI do not support Foundry.

​
Troubleshooting

Follow the restart or reopen instructions for your client in Coding Harnesses.