Tray app
The tray is the operator surface — a small control panel that lives in the system tray. It starts and stops the local server, frees your GPU on demand, manages where your data lives, and configures tools. Keep it running while you use the VS Code chat.
Starting the tray
- Use the Riverforge Tray desktop shortcut or Start Menu entry created by the installer.
- Your data lives under
%APPDATA%\Riverforgeby default — you can change this any time (see Data & Backup).
Once open, click Start Server and wait for the status log to say it’s online. If you enabled autostart during install, the tray launches silently at login and the server is ready whenever VS Code opens.
The main controls
| Button | What it does |
|---|---|
| Start Server / Stop Server | Starts or stops the local Riverforge server |
| Pause VRAM / Resume VRAM | Unloads active Ollama models to free the GPU, then warms them again |
| Status | Shows server, Ollama, model, data and tool health |
| Relay: Off / Relay: On | Toggles the outbound remote-control relay (off by default) |
Pause VRAM — the game-dev case
This is the feature built for exactly one situation: you want to play a game, run a 3D tool or compile shaders, and you need the VRAM back — but you don’t want to tear down the agent. Pause VRAM asks Ollama to unload every loaded model instantly, while the server keeps running. The next chat reloads a model when needed; Resume VRAM warms it ahead of time.
You can do the same from inside VS Code with the Pause Models / Resume Models commands, or the pause button in the chat header. Same effect, fewer windows.
The Tools tab
Tell Riverforge how your project works so the verification shortcuts just run.
| Control | Use |
|---|---|
| Tool Config | Set the default test, lint, build, format and type-check commands, and add to the allowed-command list |
| Register HTTP Tool | Add a custom HTTP-backed tool that appears beside the built-ins |
These settings power /test, /lint, run_tests and run_lint without you repeating commands. Full detail on the Tools & Commands page.
Managing your data folder
Riverforge stores chats, memory, logs, indexes, tool config and HTTP tools under one data folder. On first run, choose a stable location with enough disk space.
| Button | What it does |
|---|---|
| Choose Data Folder | Pick where chats, memory, logs and indexes live |
| Relocate Data Folder | Move or copy the data root to another drive and switch to it |
| Open Data Folder | Open the data root in Windows Explorer to inspect it |
The tray window also shows the data, brain, project-context, session and settings paths in scrollable panels. By default, project context is stored centrally under the data folder — not inside each repository — so large indexes and logs stay out of your game and app repos. The full layout and backup guide is on the Data & Backup page.
The status log
The window keeps a scrolling status log and a live indicator: model loaded, idle, busy or error. Use Status for a full snapshot of server, Ollama, model, data and tool health, and Relay Status for the remote-relay state if you’ve configured one.
Start on Windows login
The tray can register itself as a login item so the server starts silently on boot and is always ready when you open VS Code. You can enable this on the installer’s finish page, or from the tray’s startup option later.
Closing vs exiting
Closing the tray window hides it back to the system tray — it does not stop the server. To shut everything down, right-click the tray icon and choose the exit/stop option; that stops the server, unloads models and frees VRAM cleanly.
Command-line control
If you’d rather script it — for automation, a scheduled task or troubleshooting — the packaged runner does everything the tray buttons do:
PS> RiverforgeRunner.exe start # start the server
PS> RiverforgeRunner.exe status # print readiness and health
PS> RiverforgeRunner.exe pause # unload models, free VRAM
PS> RiverforgeRunner.exe stop # stop the server
Remote relay (optional)
Remote relay lets you reach the agent from a phone or other device without exposing your PC to the internet. Your home PC makes an outbound connection to a trusted relay node; the relay forwards authenticated requests back down that connection. There’s no port forwarding and no public listener on 127.0.0.1:8765.
It’s off by default and only connects once you’ve set a relay address and token. Remote requests are read-only by default; allowing remote edits is an explicit opt-in, and your normal approval still applies. Use the tray’s Relay: On / Off and Relay Status to open it only when you need it.