Skip to main content
FireConnect routes Cursor through Fireworks AI models via Cursor’s OpenAI-compatible BYOK path. See the FireConnect overview for install and CLI basics.
Change models: quit Cursor → fireconnect cursor on --model <id>. See Models.

​
Prerequisites

Quit Cursor before on or off. FireConnect writes Cursor’s SQLite settings database; the IDE must be fully closed. status is read-only while Cursor is running.
FireRouter on Cursor needs workspace BYOK (no local Anthropic key). FireConnect rejects --model firerouter with Fire Pass (fpk_...) on every harness; use an fw_... key, then fireconnect cursor on --model firerouter. See FireRouter.

​
Enable Fireworks routing

Cursor stores AI settings in a SQLite database (state.vscdb). Fully quit Cursor before running commands that write to it (for example, Cmd+Q on macOS or close all Cursor windows on Linux). Otherwise Cursor’s in-memory state can overwrite FireConnect’s changes. In an interactive terminal, if Cursor is still running FireConnect asks you to quit it and press Enter to continue. Pass --force to write anyway without waiting.
Or pass the key once:
cursor on sets every mode that already exists in modelConfig to the default Fireworks model (non-destructive: it won’t create mode entries that aren’t already there) and registers the preferred serverless catalog in the picker. Quit and reopen Cursor for the change to take effect, then open the model picker and choose a Fireworks model.

​
Browse and pick models

Browse the global catalog, then switch models with on:
fireconnect model list and status are read-only and work while Cursor is running. Commands that write to state.vscdb (on, off) require Cursor to be quit first. Short model IDs are expanded to full Fireworks paths automatically. Cursor modes include composer (default), cmd-k, background-composer, composer-ensemble, plan-execution, spec, deep-search, and quick-agent. Run fireconnect cursor status to see the current model for each mode.
Cursor enforces an allowlist on the server side. Not every Fireworks model appears in the picker even after you add it. Models such as GLM 5.2 and Kimi K2.6 are known to work; if a model is blocked, Cursor shows an error when you select it.
While FireConnect is on, only Fireworks models in your picker work. Cursor subscription models, Opus modes, and other built-in models won’t respond. Run fireconnect cursor off to restore built-in Cursor models.

​
What gets written

FireConnect writes Cursor’s BYOK OpenAI settings in the local SQLite state database at state.vscdb: Platform paths for state.vscdb: FireConnect snapshots your previous Cursor auth state under ~/.fireconnect/cursor/ before the first change. Running fireconnect cursor off restores it. off only removes models FireConnect registered; your own custom models are preserved.

​
Cursor feature coverage

FireConnect configures Cursor’s OpenAI BYOK path. Features that route through that path can use Fireworks models. Some Cursor features (for example, Composer, inline edit, and autocomplete) may still use Cursor’s own backend depending on your plan and Cursor version. Test the workflows you care about after enabling.

​
Using Fire Pass

Use your fpk_... key during login or with --api-key:
Fire Pass keys default to kimi-fast-latest.

​
CLI reference

Run fireconnect cursor help for all options, including --db-path (explicit state.vscdb path) and --force (write even if Cursor appears to be running; not recommended).

​
Turn off Fireworks routing

Quit Cursor, then run:
This restores your previous Cursor auth state from the backup in ~/.fireconnect/cursor/. Quit and reopen Cursor for full effect.

​
Manual setup

You can also configure Cursor without FireConnect:
  1. In Cursor settings, add a Custom Model with a Fireworks model ID (for example, accounts/fireworks/models/glm-5p2 or a short alias like glm-fast-latest).
  2. Set Override OpenAI Base URL to https://api.fireworks.ai/inference/v1.
  3. Paste your Fireworks or Fire Pass API key.
FireConnect automates these steps and makes it easy to swap models from the terminal.

​
Fireworks on Microsoft Foundry

Cursor supports Fireworks on Microsoft Foundry (CLI: --provider azure or on --azure). See Microsoft Foundry in FireConnect and the portal setup guide.
Foundry routing requires a standard Azure API key. Fire Pass keys (fpk_...) are not supported. Quit Cursor before on or off.
One-off routing without changing global config:
Pass your Foundry model with --model (for example, FW-GLM-5.2), not a Fireworks serverless short ID. FireConnect points Cursor’s OpenAI BYOK override at your Foundry OpenAI-compatible endpoint and registers the deployment in the model picker. fireconnect model list only browses the Fireworks serverless catalog on the direct gateway path. To switch back to the Fireworks gateway:
See Turn off Foundry routing for off and global config behavior.

​
Source

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