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.

​
Prerequisites

Azure routing not implemented yet for Cursor. fireconnect cursor on always configures direct Fireworks (https://api.fireworks.ai/inference/v1), even when global config has --provider azure or you pass --azure. See Microsoft Foundry in FireConnect.

​
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). Use cursor model select --mode <mode> to point an individual mode at a different model. Quit and reopen Cursor for the change to take effect, then open the model picker and choose a Fireworks model.

​
Browse and pick models

model list and status are read-only and work while Cursor is running. Commands that write to state.vscdb require Cursor to be quit first.
Short model IDs are expanded to full Fireworks paths automatically. Cursor modes you can pass to --mode 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.

​
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 glm-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 add or swap models from the terminal.

​
Source

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