How to use Roo Code with a self-hosted model on a cloud GPU
Rent a GPU on Deploy, then open a terminal in JupyterLab or over SSH and serve an open-weight model there. Install Roo Code the way the official install page prints, and set the documented OpenAI Compatible, Ollama, or LM Studio provider to the forwarded address of that server. Copy anything the agent wrote on the instance before you destroy it. Destroying the instance deletes its disk and stops the server.
An agent can rent the GPU and serve the model the editor calls. The agent pages are GPU compute for AI agents and the agent integration guide. Those pages are how an agent discovers a GPU and runs work on it. This page is the Roo Code setting. An editor client whose install page is the VS Code extension is How to use Cline with a self-hosted model on a cloud GPU. An editor whose config.yaml sets an apiBase for chat, autocomplete, and embeddings is How to use Continue with a self-hosted model on a cloud GPU. A terminal client whose OpenAI-compatible page sets OPENAI_API_BASE is How to use Aider with a self-hosted model on a cloud GPU. A local agent UI whose Docker command publishes port 3000 is How to run OpenHands with a self-hosted model on a cloud GPU. A coding-agent server whose card prints --tool-call-parser qwen3_coder is How to serve Qwen3-Coder for coding agents on a cloud GPU. A general vLLM install is How to serve an OpenAI-compatible API with vLLM on a cloud GPU. Ollama’s install and port 11434 are How to run Ollama on a cloud GPU. A TurboMind server whose lmdeploy serve api_server listens on port 23333 is How to serve LLMs with LMDeploy on a cloud GPU. A hosted-gateway setup whose Base URL is https://cloudgpu.app/v1 is the Roo Code API post. This page does not use that URL. How published weight sizes are read: How much VRAM do I need to run an LLM?. The tunnel is How to use SSH port forwarding on a cloud GPU. If the server runs out of memory: How to fix CUDA out of memory on a cloud GPU. If the server raises no kernel image on an RTX 5090: How to fix CUDA no kernel image on an RTX 5090.
The Roo Code LICENSE file begins with “Apache License Version 2.0, January 2004”. The appendix copyright line is “Copyright 2025 Roo Code, Inc.” The README license line is “Apache 2.0 © 2026 Roo Code, Inc.” The GitHub license field on 2026-11-14 was Apache-2.0. That license text is not a CloudGPU price.
The steps below are copied from the upstream files named in each step. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked. The provider pages also name other providers. This page does not list those names.
What the README says about the extension
The README at commit b867ec9145750d0ae1ff7f02d35406e9bf2a0b16 (committed 2026-05-15T18:04:45Z) says “The Roo Code Extension was shut down on May 15th.” The same paragraph names Zoo Code and Cline as places to look. The repository pushed_at on 2026-11-14 was 2026-05-15T18:08:47Z. The install page and the provider pages at that commit still print the steps below. This page did not install the extension.
The README marketplace badge uses the item name RooVeterinaryInc.roo-cline. A request to https://marketplace.visualstudio.com/items?itemName=RooVeterinaryInc.roo-cline returned HTTP 404 on 2026-11-14. https://docs.roocode.com/providers/openai-compatible returned HTTP 301 to https://roocodeinc.github.io/Roo-Code/providers/openai-compatible the same day. The live install page at https://roocodeinc.github.io/Roo-Code/getting-started/installing/ still contained RooVeterinaryInc, “Install from VSIX”, and “VS Code version 1.84.0”.
Choosing a GPU
The install page, the OpenAI Compatible page, and the LM Studio page print no GPU GB figure. The Ollama page troubleshooting Modelfile prints these comments:
# 16384 for ~8GB VRAM
# 32768 for ~16GB VRAM
# 65536 for ~24GB+ VRAM
Those lines are comments on a placeholder FROM line. This page does not treat them as a measured fit and does not mark a catalog row as enough. The same page’s earlier example sets num_ctx to 32768 on qwen2.5-coder:32b and does not attach those comments to that tag.
GET /v1/public/compute-capabilities does not list GPUs. It returns feature flags. The memory column is vram_gb from gpu-prices.json, updated_at 2026-11-14T03:12:17.736Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-14T03:12:18.737Z) listed the same names and the same vram_gb values, plus Ascend 910B at 64, which is not in gpu-prices.json and is not listed here. The catalog response also includes an availability field. This page does not copy it. Every prices row has gpu_count 1. Every catalog row has min_gpu_count 1. Which listing you can start is on Deploy.
The catalog template list had no Roo Code template. It did have Ollama templates (jupyterlab-ollama, ollama-qwen3, ollama). Those templates are a different install. This page does not treat them as Roo Code.
| Catalog name | vram_gb |
|---|---|
| RTX 2080 | 8 |
| RTX 3080 | 10 |
| RTX 2080 Ti | 11 |
| RTX 3080 Ti | 12 |
| V100 16G | 16 |
| A30 | 24 |
| RTX 3090 | 24 |
| RTX 4090 | 24 |
| RTX 4090D | 24 |
| RTX 4080 SUPER 32G | 32 |
| RTX 5090 | 32 |
| RTX 5090D | 32 |
| V100 32G | 32 |
| A100 40G | 40 |
| A800 40G | 40 |
| A40 | 48 |
| RTX 4090 48G | 48 |
| RTX 4090D 48G | 48 |
| RTX A6000 | 48 |
| NVIDIA RTX PRO 5000 72G | 72 |
| RTX PRO 5000 72G | 72 |
| A100 80G | 80 |
| A800 80G | 80 |
| H800 | 80 |
| RTX 6000D | 83 |
| H20 96G | 96 |
| RTX PRO 6000 96G | 96 |
| RTX PRO 6000 Max-Q | 96 |
| H20-3e 141G | 141 |
Install Roo Code
The install page’s VS Code steps, copied. Open VS Code. Open Extensions with the Activity Bar icon or Ctrl+Shift+X on Windows and Linux, or Cmd+Shift+X on macOS. Search for “Roo Code”. Select “Roo Code” by RooVeterinaryInc and click Install. Reload VS Code if prompted. The page says the Roo Code icon then appears in the Activity Bar.
The same page also prints an Open VSX route for editors without Marketplace access, and a manual VSIX route: download the .vsix from the GitHub releases page, then Extensions, the “…” menu, “Install from VSIX…”. The troubleshooting list says to verify VS Code version 1.84.0 or later. This page did not install from either registry and did not download a VSIX. Roo Code itself is the editor extension. The weights are whatever model server you start on the instance.
Modes and the context window
The README lists Code, Architect, Ask, Debug, and Custom modes. The using-modes page also documents Orchestrator, which it also calls Boomerang Mode. Switch with the dropdown left of the chat input, with /architect, /ask, /debug, /code, or /orchestrator, or with Ctrl + . on Windows and Linux and ⌘ + . on macOS. Code and Debug have tool groups read, edit, command, and mcp. Ask has read and mcp. Architect has read, mcp, and restricted edit for markdown files only. Orchestrator has no direct tool access and uses the new_task tool. A note on that page assigns example models to modes. This page does not copy those names.
The OpenAI Compatible page’s Model Configuration list includes Max Output Tokens and Context Window. It also includes price fields. This page does not copy those price fields. The page does not print a default number for Context Window.
The Ollama page says Roo Code defers to the Modelfile num_ctx by default. The interactive example, copied:
ollama run qwen2.5-coder:32b
/set parameter num_ctx 32768
/save your_model_name
The Modelfile example, copied:
# Example Modelfile for reduced context
FROM qwen2.5-coder:32b
# Set context window to 32K tokens (reduced from default)
PARAMETER num_ctx 32768
# Optional: Adjust temperature for more consistent output
PARAMETER temperature 0.7
# Optional: Set repeat penalty
PARAMETER repeat_penalty 1.1
ollama create qwen-32k -f Modelfile
That 32768 is the example on that page. This page did not save the model.
Point the provider at the server
The OpenAI Compatible page says to open Roo Code settings and set API Provider to “OpenAI Compatible”, then Base URL, API Key, and Model. It says the Base URL is the provider endpoint. The page also prints one hosted API URL. This page does not copy it. For a server on the instance, Base URL is the client address that server’s guide prints, after the port is forwarded. How to use SSH port forwarding on a cloud GPU is how a program on your computer reaches a port on the instance. This page does not add an ssh command.
The same page says Roo Code uses native tool calling only, and that there is no XML-based fallback. The requirements line is: “If a model does not support native tool calling, it cannot be used with Roo Code.” The troubleshooting list repeats that for tool-calling errors, and it also lists “Invalid API Key”, “Model Not Found”, and connection errors when the Base URL is wrong.
The Ollama page’s provider steps, copied. Select “ollama” as the API Provider. Enter the model tag or the saved name, for example your_model_name. The optional base URL defaults to http://localhost:11434. An API key is optional. The page says the “ollama” provider uses Ollama’s native API by default, and that an OpenAI-compatible /v1 handler also exists but is not required for typical setups. The download lines, copied. The first command’s model tag is an angle-bracket placeholder. It stays in the fence so the renderer does not treat it as HTML.
ollama pull <model_name>
ollama pull qwen2.5-coder:32b
The out-of-memory section on that page says the first request can fail when no model instance is already running, because Ollama may allocate a larger context window than expected. The fixes it prints are to preload with ollama run, to pin num_ctx, to use a smaller quant or a smaller parameter size, and to run ollama ps then ollama stop. This page does not copy a duration from that section.
The LM Studio page says to choose “LM Studio” as the API Provider, enter the file name of the loaded model, and that the default base URL is http://localhost:1234. The example file name on that page is codellama-7b.Q4_0.gguf. The local server in LM Studio has to be started before Roo Code connects. A troubleshooting line says a “Please check the LM Studio developer logs to debug what went wrong” error may need a context-length change in LM Studio. That line prints no token count.
The local-models page says Roo Code supports Ollama and LM Studio for a model on your own machine. One connection error it quotes is “No connection could be made because the target machine actively refused it”, which it attributes to the server not running or to a different port than the Base URL.
When it fails
CUDA out of memory. The only GPU-sized numbers this page found are the Ollama page’s Modelfile comments, ~8GB, ~16GB, and ~24GB+ VRAM. Those are comments. If the model server raises CUDA out of memory, use how to fix CUDA out of memory on a cloud GPU. This page does not pick a smaller model id for you.
“CUDA error: no kernel image is available for execution on the device” on an RTX 5090. The Roo Code pages this page read do not pin a PyTorch build and do not name compute capability 12.0. The server you start might. The RTX 5090 guide says that card is compute capability 12.0 (sm_120) and that PyTorch 2.7 and later, on a build for CUDA 12.8 or newer, is the line that announced support. This page did not install a torch wheel and did not re-test a server on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only.
Tool calling. The OpenAI Compatible page says a model without native tool calling cannot be used with Roo Code. This page did not send a tools request.
What to copy off, then destroy
The Roo Code pages do not name an output directory on the GPU. Chat transcripts stay in the editor. Any file the agent writes on the instance is on that disk. Copy it before you destroy the instance. Save your work before you destroy a cloud GPU is the checklist. The terms say data on local instance storage is permanently deleted when the instance is destroyed. The server stops with the instance.
The pricing page FAQ states the billing rule in these words:
Per-minute options have a 1-minute minimum; unused prepaid minutes are refunded when you destroy the instance. Some configurations use whole-hour billing with a 1-hour minimum; unused time within a charged hour is not refunded on voluntary stop. The billing period, minimum charge and initial prepayment are shown before confirmation. No subscription.
Prices are not copied here.
Frequently asked questions
What memory figures does the Ollama page print?
The install page, the OpenAI Compatible page, and the LM Studio page print no GPU GB figure. The Ollama troubleshooting Modelfile comments say 16384 for ~8GB VRAM, 32768 for ~16GB VRAM, and 65536 for ~24GB+ VRAM. Those lines are comments on a placeholder FROM line. This page does not mark a catalog row as enough. Which listing you can start is on https://cloudgpu.app/deploy.
Which install steps does the install page print, and what does the README say about the extension?
The install page VS Code steps are Extensions, search Roo Code, select Roo Code by RooVeterinaryInc, Install. It also prints Open VSX and Install from VSIX. The troubleshooting list says VS Code version 1.84.0 or later. The README says the Roo Code Extension was shut down on May 15th and names Zoo Code and Cline. The marketplace item URL for RooVeterinaryInc.roo-cline returned HTTP 404 on 2026-11-14. This page did not install the extension.
How do the OpenAI Compatible, Ollama, and LM Studio settings reach a server on the rented GPU?
The OpenAI Compatible fields are Base URL, API Key, and Model. Model Configuration includes Max Output Tokens and Context Window, and also includes price fields this page does not copy. The Ollama provider default base URL is http://localhost:11434, and the page says Roo defers to the Modelfile num_ctx. The documented example sets num_ctx to 32768. The LM Studio default base URL is http://localhost:1234. The SSH port-forwarding guide is how the editor machine reaches the instance port. This page did not open a tunnel. The OpenAI Compatible page says Roo Code uses native tool calling only, and that a model without native tool calling cannot be used with Roo Code.
What license does the file print?
The Roo Code LICENSE file begins with Apache License Version 2.0, January 2004. The appendix copyright line is Copyright 2025 Roo Code, Inc. The README license line is Apache 2.0 © 2026 Roo Code, Inc. The GitHub license field on 2026-11-14 was Apache-2.0. That license text is not a CloudGPU price.
Sources (read 2026-11-14, Asia/Shanghai)
- RooCodeInc/Roo-Code README (shutdown sentence, modes Code / Architect / Ask / Debug / Custom, marketplace badge
RooVeterinaryInc.roo-cline, Apache 2.0 line): commitb867ec9145750d0ae1ff7f02d35406e9bf2a0b16, committed 2026-05-15T18:04:45Z. GitHub license fieldApache-2.0.pushed_at2026-05-15T18:08:47Z.LICENSEappendix: Copyright 2025 Roo Code, Inc. apps/docs/docs/getting-started/installing.mdx(VS Code search, Open VSX, Install from VSIX, VS Code 1.84.0): same commit. Live URL https://roocodeinc.github.io/Roo-Code/getting-started/installing/ contained those strings on 2026-11-14. Marketplace item URL returned HTTP 404 the same dayapps/docs/docs/providers/openai-compatible.md(API Provider “OpenAI Compatible”, Base URL, API Key, Model, Max Output Tokens, Context Window, native tool calling): same commit. That page also names other providers and one hosted API URL. This page does not list them.https://docs.roocode.com/providers/openai-compatiblereturned HTTP 301 to the github.io URLapps/docs/docs/providers/ollama.md(ollama pull qwen2.5-coder:32b,num_ctx32768, defaulthttp://localhost:11434, VRAM comments,ollama ps/ollama stop): same commitapps/docs/docs/providers/lmstudio.md(provider “LM Studio”, examplecodellama-7b.Q4_0.gguf, defaulthttp://localhost:1234): same commitapps/docs/docs/basic-usage/using-modes.md(Orchestrator, slash commands, tool groups): same commit. The sticky-models note names models this page does not copyapps/docs/docs/advanced-usage/local-models.md(Ollama and LM Studio, connection-refused sentence): same commit- Public GPU memory: https://cloudgpu.app/gpu-prices.json (
updated_at2026-11-14T03:12:17.736Z) and https://cloudgpu.app/v1/public/gpu-catalog (generated_at2026-11-14T03:12:18.737Z).GET /v1/public/compute-capabilitiesreturned feature flags and no GPU list. The catalog had no Roo Code template
Cards you can start are on the Deploy page. Prices are on https://cloudgpu.app/pricing and are not repeated here.
Last updated: 2026-11-14 (Asia/Shanghai).