Frequently asked questions
The short answers. Each links onward to the page with the full story.
Privacy & cost
Does any of my code leave my machine?
No. Every model call, edit and memory write runs locally through Ollama. There’s no server to send anything to — and you can prove it: run netstat -ano | findstr :8765 and you’ll see a local-only listener, no calls out.
Do I need an internet connection?
Not for the core loop. Models, embeddings, search index, tool execution and memory are all local. The web tools are explicitly off by default and only fire when you ask for research. It works on a plane and during an outage.
Is it free? Where do I get it?
Riverforge is free and currently in private beta — the public Windows download lands on the website soon. See Installation.
Hardware
What hardware do I need?
Minimum: Windows 10/11, 16 GB RAM, an NVIDIA GPU with 8 GB VRAM, and 40–80 GB of disk. Recommended: 32 GB RAM and an RTX 3070 or better. Full table on Installation.
Can I run it without a GPU?
Yes — CPU-only works, you just accept much slower responses. Riverforge was designed low-end first and scales up with your hardware.
Can I use it while gaming or doing 3D work?
Yes. Click Pause VRAM in the tray (or run the Pause Models command) and Ollama unloads its models instantly while the server stays up. The next chat reloads a model when needed. See Tray App.
Models
Which model does it use?
Whichever local model you choose. The installer offers several benchmarked options through Ollama — Gemma 4 Turbo is pre-selected as the fast pick for ~8 GB GPUs, with more capable choices like Qwen 3.5 9B or Gemma 4 26B if you have the VRAM. You can switch any time. See Choosing your model.
Can I use a different or bigger model?
Yes. Point it at any Ollama model — Gemma, Qwen, Llama, Phi, your own fine-tune — from the Models panel or the API. Riverforge adapts how it prompts and runs to suit it. See Models & Hardware.
Daily use
What’s the difference between Chat, Ask and Agent?
Chat is conversation. Ask is grounded, read-only help — explanations and audits that never touch files. Agent does the full job — it inspects, edits, runs tests and handles multi-step work. See Chat Modes.
Will it edit my files without asking?
Only if you let it. With the Confirm setting, risky tools pause for your approval. With Autopilot, they run — but diff review, verification and rollback still apply. You can flip the setting live, mid-run. See Approvals & Undo.
How do I undo a change?
Use the Undo all button above the composer to reverse a run’s changes, or the Undo button on any past response. For shipped work, a git commit per verified change is still the safest audit trail.
How do I give it specific files to work with?
Attach them — the + button, or drag files/folders from the VS Code or Windows Explorer into the chat. Folders are expanded automatically, skipping generated directories. See Attachments.
Does it actually run my tests?
Yes. After changing files, the agent is required to run your tests or lint before it can finish — and if they fail, it repairs and re-runs. Tell it your project’s commands once in the tray under Tool Config. See Tools & Commands.
Memory
Does it remember things between sessions?
Yes. It keeps a brain that grows over months — your preferences and style, per-project decisions, and an assistant identity it earns. Facts strengthen when used and gently fade when not; nothing is ever hard-deleted. See Memory & Identity.
How do I make it remember or forget something?
Just say so — “remember that I prefer pytest and small commits” or “forget what you stored about the old auth flow.” No special syntax. You can also browse, edit or delete any memory in the Memory Visualizer.
Can I back up my memory?
Yes — your memory lives in your Riverforge data folder, so backing up that folder backs up everything. To move just your memory to another machine, use the export. See Data & Backup.
Setup & problems
The chat says the server is offline.
Open the Riverforge Tray, click Start Server, and wait for the status to go green. The VS Code extension reconnects automatically once the server is up.
The VS Code command is missing after install.
Install the bundled VSIX manually into the window you actually use: Extensions view → … → Install from VSIX…, then reload. The CLI can install into a different profile. See Installation.
Responses are slow or look cut off.
The first response after startup includes model warmup — use Resume VRAM to warm ahead. Truncated answers are almost always Ollama’s 2048-token default context; set OLLAMA_CONTEXT_LENGTH=16384 and restart Ollama. More on Troubleshooting.