Overview

Dashboard

Live health, account, and traffic at a glance.

Requests, last 24 hours

Requests by model

Recent activity

Activity

Request log

Every request this gateway process has handled, live (in-memory only - resets on restart).

Sessions

Active conversations

Cached Cursor agents keeping multi-turn conversations alive. Evicting one just means that conversation's next message starts a fresh agent.

Models

Model catalog

Every model available to the configured Cursor account. Click a model to make it the gateway's default.

Settings

Configuration

Every configurable option on this gateway - nothing is left to a config file only. Changes apply live; a few need a restart, which you can also trigger right here.

Cursor API key

The credential this gateway uses to talk to Cursor on your behalf.

Key mode

Whether every client shares this one key, or supplies its own.

Runtime

Where Cursor agents actually execute.

Agent conversation mode

"Plan" only reasons/plans; it never edits files or runs commands.

Default model

Used whenever a client omits "model" or requests one you don't have access to. Browse the full catalog under Models.

Sessions

How multi-turn conversations reuse a live Cursor agent instead of replaying history.

Enable sessions Auto-detect continuations (for clients that resend full history)

Session idle timeout

Minutes of inactivity before a cached agent is evicted.

Max cached agents

Oldest-used sessions are evicted first once this limit is reached.

Reasoning

Stream the model's chain-of-thought as an OpenAI-style reasoning_content field.

Include reasoning in responses

Tool-calling bridge

Translates OpenAI tools[] function-calling into Cursor's native tool system (local runtime only).

Enable OpenAI tool-calling bridge

Tool bridge mode

"Hold" keeps one Cursor run alive across a whole tool loop (one metered request, captures parallel calls). "Cancel" is the legacy one-run-per-step behavior.

Tool result timeout

Hold mode only: seconds a held run waits for the client's tool result before it's torn down (freeing the agent and its concurrency slot).

Max concurrent runs

Total Cursor agent runs allowed in flight at once, across all clients.

Request timeout

Seconds before an in-progress run is cancelled and reported as a timeout.

Rate limit — max requests

Per client (by API key/IP), per window.

Rate limit window

Seconds per window. Needs a restart to take effect - use the button below once you save.

Max request body size (MB)

1–1024. Oversized bodies get HTTP 413 (clients like Hermes compress their history and continue). Needs a restart to take effect.

Changing host/port reconnects the running server on the new address — this page follows automatically.

Host

Network interface to listen on. 0.0.0.0 to accept connections from other devices.

Port

If already in use on startup, the gateway automatically tries the next few ports instead.

CORS origin

Allowed origins for browser-based clients. * for any.

Auto-open browser

Open this dashboard automatically whenever the gateway process starts.

Open on startup

Admin key

Protects this dashboard, its API, and (if you choose) the OpenAI-compatible endpoints. Generate one, or set your own.

New admin key — save it now, it will not be shown again:

Remote dashboard access

By default the dashboard/API only accept requests from this machine, regardless of the admin key.

Allow from other devices on the network (not recommended)

Log level

Verbosity of server logs.

Pretty-print logs

Human-readable console output instead of raw JSON lines.

Pretty-print

Workdir root

Where each session's isolated scratch directory is created. Applies live to new sessions; existing cached sessions keep the directory they already have.

Node environment

Informational only - nothing in this gateway currently branches on it.

Restart gateway

Fully applies any pending "restart required" change (like the rate limit window above) by respawning the process. Takes a few seconds; this page reconnects automatically.

Export

Download the full current configuration (including secrets) as JSON, for your own records or to restore later.

Import

Restore settings from a previously exported file. Unrecognized or secret-looking fields are safely skipped and reported below.

Click to choose a configuration JSON file…

System info

Read-only diagnostics about the running process.

Chat

Try the gateway

A real conversation through this gateway, using your configured Cursor account — no code needed.

Start a conversation below.

Connect

Wire it up

Point any OpenAI-compatible client at this gateway, or use the CLI for scripted/automated setups.

Connect any OpenAI-compatible tool

Point it at this base URL and API key. Pick a model to fill the snippets below.

Base URL

API key

Model for snippets

Let's get you connected

You're all set

Sign in