How to use Aider with a self-hosted model on a cloud GPU

Rent a GPU on Deploy, then open a terminal in JupyterLab or over SSH. Install Aider with the command on the official install page, serve an open-weight model with vLLM, SGLang, or Ollama on that machine, and point Aider at the endpoint the OpenAI-compatible page or the Ollama page documents. Copy the git repo Aider commits into before you destroy the instance, because destroying the instance deletes its disk and stops the server.

An agent can rent the GPU, serve a model, and run Aider against that endpoint. 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 Aider command. 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 local agent UI whose Docker command publishes port 3000 is How to run OpenHands with a self-hosted model on a cloud GPU. A general vLLM install is How to serve an OpenAI-compatible API with vLLM on a cloud GPU. A general SGLang install is How to run SGLang on a cloud GPU. Ollama’s install and port 11434 are How to run Ollama on a cloud GPU. CloudGPU’s hosted gateway is a different base URL: How do I use an OpenAI-compatible API for open models?. How published weight sizes are read: How much VRAM do I need to run an LLM?. If the process 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 tunnel is How to use SSH port forwarding on a cloud GPU. An editor client whose install page is the VS Code extension, and whose OpenAI Compatible setting takes the forwarded Base URL, is How to use Cline with a self-hosted model on a cloud GPU.

The Aider LICENSE.txt file begins with “Apache License Version 2.0, January 2004”. The appendix copyright line is the unfilled placeholder Copyright [yyyy] [name of copyright owner]. The GitHub license field on 2026-11-14 was Apache-2.0. That license text is not a CloudGPU price.

The commands below are copied from the upstream pages named in each step. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked. The install page’s start block also prints --model lines for hosted models. This page does not copy those lines.

What the Aider pages print about local models

The connecting-to-LLMs page (fetched 2026-11-14) says Aider can work with local models, for example using Ollama, and that it can also access local models that provide an OpenAI compatible API. The same page’s other sections name hosted models. This page does not copy those names.

The OpenAI-compatible page says Aider can connect to any LLM which is accessible via an OpenAI compatible API endpoint. Its model line is aider --model openai/<model-name>. The prefix openai/ is the prefix that page prints. It is not a model id. The page says to see the model warnings section for models aider is not familiar with.

The model warnings page prints this warning:

Model foobar: Unknown context window size and costs, using sane defaults.

It says this means aider does not know the context window size and token costs for that model, and that you can probably ignore the unknown context window size and token costs warning. It points at the advanced model settings page for how to remove the warning. One sentence on that page uses a word this page does not copy when it describes the default window. This page does not copy that sentence.

The edit formats page says different models work better or worse with different edit formats, and that you can force a format with --edit-format. The formats that page documents are whole, diff, diff-fenced, udiff, editor-diff, and editor-whole. It does not name a format for a local model. The paragraphs for diff-fenced and udiff name model families. This page does not copy those names.

The Ollama page says ollama_chat/ is recommended over ollama/. It says Ollama uses a 2k context window by default, which is very small for working with aider, and that it silently discards context that exceeds the window. Its serve line sets OLLAMA_CONTEXT_LENGTH=8192. It also says that by default aider sets Ollama’s context window to be large enough for each request plus 8k tokens for the reply. A fixed window is an .aider.model.settings.yml entry, copied below.

The install page, the OpenAI-compatible page, the Ollama page, the edit formats page, and the warnings page print no GPU GB figure. The memory is the model server’s, and those servers are the other guides linked above. This page does not mark a catalog row as enough for Aider.

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 2026-11-14T00:03:37.866Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-14T00:03:37.294Z) 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 Aider template. It did list Ollama, JupyterLab + Ollama, and Ollama + Qwen3. Those names are template titles. They are not the model string on the Aider Ollama page.

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 Aider

On Deploy, choose Raw GPU. The install page says: if you already have python 3.8-3.13 installed, you can get started quickly like this. Copied, without the hosted --model lines that follow cd on that page:

python -m pip install aider-install
aider-install

The page says this installs aider in its own separate python environment, and that aider-install will also install a separate version of python 3.12 to use with aider if needed.

The same page’s uv section, copied:

python -m pip install uv  # If you need to install uv
uv tool install --force --python python3.12 --with pip aider-chat@latest

It says this installs uv using your existing python version 3.8-3.13, and that uv will automatically install a separate python 3.12 to use with aider if needed.

The pip section says you can use pip to install aider with python versions 3.9-3.12, and that you should consider a virtual environment. Copied:

python -m pip install -U --upgrade-strategy only-if-needed aider-chat

It says that in some environments you may get “aider command not found” errors, and that you can try python -m aider. The pipx section says pipx install aider-chat with python versions 3.9-3.12. The Linux one-liner is curl -LsSf https://aider.chat/install.sh | sh. This page did not run the installer.

The docker page says Aider is available as two images, paulgauthier/aider and paulgauthier/aider-full. Its docker run lines pass --openai-api-key. This page follows the OpenAI-compatible page’s environment variables instead of those lines.

The Aider pages this guide read do not print hf download or huggingface-cli. Weight download is the model server’s guide. How to download a Hugging Face model prints pip install -U "huggingface_hub" and hf download.

Point Aider at the endpoint

The OpenAI-compatible page’s configuration, copied. <endpoint> and <key> are the placeholders that page prints. <model-name> is the placeholder that page prints. This page does not fill them in.

# Mac/Linux:
export OPENAI_API_BASE=<endpoint>
export OPENAI_API_KEY=<key>

# Windows:
setx OPENAI_API_BASE <endpoint>
setx OPENAI_API_KEY <key>
# ... restart shell after setx commands

# Change directory into your codebase
cd /to/your/project

# Prefix the model name with openai/
aider --model openai/<model-name>

The page does not print /v1 and does not print a port. A vLLM server’s default listen address is a different document: the vLLM guide says the quickstart server starts at http://localhost:8000. An SGLang server’s port is the SGLang guide. Put the endpoint you actually started into OPENAI_API_BASE. This page does not merge those server ports into the Aider command.

You can run that aider command on the GPU box, in the repo on the instance. You can also run it on your computer after a tunnel to the server port. The Aider pages do not print an ssh command. 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 Ollama page, copied. <model> and <api-key> are the placeholders that page prints.

export OLLAMA_API_BASE=http://127.0.0.1:11434 # Mac/Linux
setx   OLLAMA_API_BASE http://127.0.0.1:11434 # Windows, restart shell after setx

# Pull the model
ollama pull <model>

# Start your ollama server, increasing the context window to 8k tokens
OLLAMA_CONTEXT_LENGTH=8192 ollama serve

# In another terminal window, change directory into your codebase
cd /to/your/project

aider --model ollama_chat/<model>

The page’s API-key block, if the server requires one:

export OLLAMA_API_KEY=<api-key> # Mac/Linux
setx   OLLAMA_API_KEY <api-key> # Windows, restart shell after setx

The page’s fixed context-window file, copied. The name value uses the prefix ollama/. The launch line above uses ollama_chat/. This page does not change either string.

- name: ollama/qwen2.5-coder:32b-instruct-fp16
  extra_params:
    num_ctx: 65536

Edit format

The edit formats page says you can force a format with --edit-format. It does not print a value for a local model. This page does not add one. The Ollama page’s settings file is the local-model setting that page prints. It sets num_ctx and does not set an edit format.

Browser UI

The browser page says to use --browser to launch the browser version of aider. Its command block, copied except that this page does not repeat the hosted-model sentence above the block:

python -m pip install -U aider-chat

export OPENAI_API_KEY=<key> # Mac/Linux
setx   OPENAI_API_KEY <key> # Windows, restart shell after setx

aider --browser

The block does not print a port, and it does not print OPENAI_API_BASE. This page does not invent a port and does not merge the OpenAI-compatible page’s OPENAI_API_BASE into this block. If the process prints a port, the SSH port-forwarding guide is how a browser on your computer reaches it. This page did not launch --browser.

When it fails

Unknown context window. The warnings page prints the Model foobar warning above and says you can probably ignore it. It points at advanced model settings to remove it. This page did not write a settings file beyond the Ollama page’s num_ctx example.

Edits that do not apply. The connecting-to-LLMs page says that if you see the model returning code, but aider is not able to edit your files and commit the changes, this is usually because the model is not capable of properly returning code edits. The edit formats page says you can force --edit-format. It does not name the format for a local model. This page does not pick one.

CUDA out of memory. The Aider pages this guide read print no GPU GB figure. The memory error comes from the model server. If that process raises CUDA out of memory, use how to fix CUDA out of memory on a cloud GPU.

“CUDA error: no kernel image is available for execution on the device” on an RTX 5090. The Aider pages do not pin a PyTorch build and do not name compute capability 12.0. The server guides are the ones that install a GPU stack. The RTX 5090 guide says that card is compute capability 12.0 (sm_120). This page did not re-test a server on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only.

What to copy off, then destroy

Aider commits in the git repo you started it in. The install page says to change directory into your codebase. Copy that repo, and any file you wrote on the instance, before you destroy it. 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

How much VRAM do the Aider pages print?

The install page, the OpenAI-compatible page, the Ollama page, the edit formats page, and the model warnings page print no GPU GB figure. The Ollama page says Ollama uses a 2k context window by default, sets OLLAMA_CONTEXT_LENGTH=8192 on its serve line, and says aider sizes the window for the request plus 8k tokens for the reply. This page does not mark a catalog row as enough. Which listing you can start is on https://cloudgpu.app/deploy.

Which command points Aider at a self-hosted endpoint?

The OpenAI-compatible page sets OPENAI_API_BASE to <endpoint> and OPENAI_API_KEY to <key>, then runs aider --model openai/<model-name>. The Ollama page sets OLLAMA_API_BASE to http://127.0.0.1:11434, runs ollama pull <model>, starts the server with OLLAMA_CONTEXT_LENGTH=8192 ollama serve, and runs aider --model ollama_chat/<model>. It says ollama_chat/ is recommended over ollama/. Its settings example names ollama/qwen2.5-coder:32b-instruct-fp16 and sets num_ctx to 65536. This page did not run them.

Does Aider ship a web UI, and which port does that page print?

The browser page’s command is python -m pip install -U aider-chat, then aider --browser. The same block exports OPENAI_API_KEY. It does not print a port and does not print OPENAI_API_BASE. This page does not invent a port. The SSH port-forwarding guide is how a program on your computer reaches a port on the instance. The CLI itself is a terminal. This page did not launch the browser UI.

What license does the file print?

The LICENSE.txt file begins with Apache License Version 2.0, January 2004. The appendix copyright line is the unfilled placeholder Copyright [yyyy] [name of copyright owner]. The GitHub license field on 2026-11-14 was Apache-2.0. Repository HEAD was 5dc9490bb35f9729ef2c95d00a19ccd30c26339c, committed 2026-05-22T14:02:20Z.

Sources (read 2026-11-14, Asia/Shanghai)

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).