Manual setup (advanced)

Configure a coding harness for Fireworks by hand, without installing the Nexus CLI.

The recommended setup is the Nexus CLI, which writes your harness settings and restores the originals with nexus <harness> off. Use manual setup when you can’t install a CLI or need settings the Nexus CLI doesn’t write.

​

When to use manual setup

• You can’t install the Nexus CLI.
• You manage harness configs yourself, through dotfiles or a team rollout.
• You want Nexus Auto to use Anthropic or OpenAI models without Provider Keys or the Nexus CLI.
• You want to review every setting that changes.

​

What changes without the Nexus CLI

The harness calls the same Fireworks endpoints with your own API key. You take over the tasks that the Nexus CLI otherwise performs.

	With the Nexus CLI	With manual setup
Fireworks models in the picker	Added automatically	Set the model ID in your config
Anthropic model turns in Claude Code	Paid by your Claude login	Need an Anthropic Provider Key or an x-anthropic-api-key header
MCP tool search	Turned on	Claude Code: add ENABLE_TOOL_SEARCH. Codex: write a model catalog that sets supports_search_tool (see What the Nexus CLI writes)
Session cost status line (Claude Code)	Installed	Not available
Anthropic and OpenAI keys for Nexus Auto	Saved once with nexus auto and written into each harness that can send them	Add x-anthropic-api-key or x-openai-api-key headers to each config
Undo	nexus <harness> off	Restore your own backup
Updates	nexus update refreshes connected harnesses	Update your configs yourself

​

Before you start

1. Create an API key in Nexus settings and export it:

   export FIREWORKS_API_KEY="fw_..."
   

2. Back up the config file you’re about to edit.

3. Use these endpoints:

   API	Base URL to configure	Full endpoint	Typical clients
   Anthropic Messages	https://api.fireworks.ai/inference	https://api.fireworks.ai/inference/v1/messages	Anthropic SDK, Claude Code
   OpenAI Chat Completions	https://api.fireworks.ai/inference/v1	https://api.fireworks.ai/inference/v1/chat/completions	OpenAI SDK, Fireworks SDK, most gateways and harnesses
   OpenAI Responses	https://api.fireworks.ai/inference/v1	https://api.fireworks.ai/inference/v1/responses	OpenAI SDK, Codex

   Anthropic-style clients add /v1/messages to the base URL themselves, so their base URL ends at /inference.

Use any model ID from Open Models, such as glm-latest, or auto for Nexus Auto.

​

Set up your harness

Follow the manual setup section for your harness:

• Claude Code
• Codex
• ChatGPT desktop app
• OpenCode
• Pi
• Cursor IDE
• VS Code
• Copilot App
• Copilot CLI
• DeepSeek Harness

The Claude Agent SDK reads your Claude Code settings, so set up Claude Code first. Claude Desktop (Limited) has no manual setup and requires the Nexus CLI.

​

Use Nexus Auto with manual setup

For Nexus Auto to use an Anthropic or OpenAI model, either:

• The harness sends an x-anthropic-api-key or x-openai-api-key header, or
• Your account has Provider Keys.

Cursor IDE and the Copilot App can’t send extra headers, so they need Provider Keys. The harness setup builder generates a config with the headers included.