Skip to main content
FireConnect routes Claude Code through Fireworks AI models. See the FireConnect overview for install and CLI basics.

Prerequisites

Azure routing not implemented yet for Claude Code. fireconnect claude on always configures direct Fireworks, even when global config has --provider azure or you pass --azure. See Microsoft Foundry in FireConnect.

Enable Fireworks routing

Or pass the key once:
Restart Claude Code after enabling, then test with:
After fireconnect claude on, model select, or model reset, settings are updated immediately. To use a new model in the same session, run /model in Claude Code, start a new session, or /exit and resume with claude --resume <id>.

Using Fire Pass

Use your fpk_... key during login or with --api-key:
FireConnect detects Fire Pass keys and routes all model aliases to glm-fast-latest.

Default model mapping

AliasStandard key (fw_...)Fire Pass key (fpk_...)
mainglm-fast-latestglm-fast-latest
opusglm-fast-latestglm-fast-latest
fableglm-fast-latestglm-fast-latest
sonnetglm-5p1glm-fast-latest
haikudeepseek-v4-flashglm-fast-latest
subagentdeepseek-v4-flashglm-fast-latest
Short model IDs like glm-fast-latest are expanded to full Fireworks paths (for example, accounts/fireworks/routers/glm-fast-latest[1m]). FireConnect appends the [1m] suffix on main, opus, and fable so Claude Code enables 1M context. The subagent slot is written without [1m] because Claude Code forwards that value verbatim to the provider API.

What gets written

FireConnect writes these settings to ~/.claude/settings.json (no literal API key in the file — the key lives in the OS keychain and is fetched at runtime via apiKeyHelper):
Claude Code calls apiKeyHelper at runtime to fetch the key from the OS keychain. FireConnect saves a backup of your previous provider settings to ~/.fireconnect/claude/ so fireconnect claude off can restore them.

Browsing and picking models

fireconnect claude model list

Lists serverless models from the Fireworks API (supports_serverless=true) and merges known public platform routers (glm-latest, glm-fast-latest, glm-5p2-fast, kimi-fast-latest, kimi-latest, kimi-k2p6-turbo, and kimi-k2p7-code-fast). Every row is tagged serverless.
Resolves the key from --api-key, the OS keychain, ~/.fireconnect/config.json, or FIREWORKS_API_KEY. Fire Pass keys (fpk_...) show Fire Pass-supported routers only.

fireconnect claude model select

Interactive picker. Requires a terminal and Fireworks to be enabled.

fireconnect claude status vs fireconnect claude model list

CommandShows
fireconnect claude statusYour current provider, auth, configured alias mapping, and Fireworks serverless rates per slot
fireconnect claude model listAvailable serverless endpoints from the Fireworks API, with IN / OUT pricing where known

Claude Code pricing estimates

Claude Code’s /model picker and session cost estimates use Anthropic list prices, not Fireworks serverless rates. For example, the default glm-fast-latest mapping may show Opus-tier estimates around 5/5 / 25 per Mtok while Fireworks bills at model-specific serverless rates (often much lower). FireConnect cannot override Claude Code’s price column. Use fireconnect claude status and fireconnect claude model list for Fireworks rates, and check the billing dashboard for actual spend. FireConnect also writes ANTHROPIC_CUSTOM_MODEL_OPTION_DESCRIPTION with Fireworks rates for your main model — that text appears as the subtitle on the custom picker entry at the bottom of /model, not in the price column.

CLI reference

Run fireconnect claude help for all options.

Switch models

Or pick interactively:

Turn off Fireworks routing

This restores your previous ~/.claude/settings.json from the backup saved in ~/.fireconnect/claude/.

Uninstall

To remove FireConnect from your machine entirely (all harnesses):

Source

FireConnect is open source: github.com/fw-ai/fireconnect