How to run Open Interpreter with a self-hosted model on a cloud GPU
Rent a GPU on Deploy, then open a terminal in JupyterLab or over SSH. Install Open Interpreter with the curl command the official install page prints, serve a model on that machine, and point the local Ollama provider at the base URL the models page prints. Copy ~/.openinterpreter before you destroy the instance, because destroying the instance deletes its disk, and that directory holds the configuration, sessions, and logs.
An agent can rent the GPU and run this program on it. 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 Open Interpreter install. An agent whose Ollama example sets base_url to http://localhost:11434/v1 is How to run OpenManus with a self-hosted model on a cloud GPU. A workspace app whose Docker quickstart publishes port 3001, and whose Ollama page sets the URL to http://127.0.0.1:11434, is How to run AnythingLLM agents with a self-hosted model on a cloud GPU. A repository agent whose local-models section sets api_base to http://localhost:11434 is How to run SWE-agent with a self-hosted model on a cloud GPU. A workflow editor whose Docker install publishes port 5678, and whose Ollama credentials default the Base URL to http://localhost:11434, is How to run n8n AI agents with a self-hosted model on a cloud GPU. A CLI whose providers page sets OLLAMA_HOST to http://localhost:11434 is How to run the goose agent with a self-hosted model on a cloud GPU. A coding agent whose providers page sets package to the OpenAI-compatible provider is How to use OpenCode with a self-hosted model on a cloud GPU. A canvas whose self-host page opens http://localhost/install is How to run Dify with a self-hosted model on a cloud GPU. A canvas whose ChatOllama page names a base URL is How to run Flowise with a self-hosted model on a cloud GPU. A flow editor whose Ollama component defaults the base URL to http://127.0.0.1:11434 is How to run Langflow with a self-hosted model on a cloud GPU. A CLI whose connect help sets --base-url to http://localhost:8000/v1 for openai-compatible is How to run Letta agents with a self-hosted model on a cloud GPU. A browser chat whose serve command listens on port 8080 is How to run Open WebUI on a cloud GPU. A Python agent library whose CodeAgent can take TransformersModel or an OpenAI-compatible server is How to run smolagents with a self-hosted model on a cloud GPU. A crew whose LLM class takes base_url for Ollama or an OpenAI-compatible server is How to run CrewAI with a self-hosted model on a cloud GPU. A graph runtime whose overview example does not name a model is How to run LangGraph agents with a self-hosted model on a cloud GPU. An AgentChat library whose Ollama example sets model to llama3.2 is How to run AutoGen agents with a self-hosted model on a cloud GPU. A typed agent whose vLLM example sets base_url to http://localhost:8000/v1 is How to run Pydantic AI agents with a self-hosted model on a cloud GPU. A framework whose README sets Ollama and HuggingFaceEmbedding is How to build LlamaIndex agents with a self-hosted model on a cloud GPU. A browser agent whose models page sets base_url to http://localhost:11434/v1 is How to run browser-use with a self-hosted model on a cloud GPU. A pipeline whose OllamaChatGenerator sets url to http://localhost:11434 is How to run Haystack agents with a self-hosted model on a cloud GPU. A program whose language-models page sets dspy.LM to ollama_chat/llama3.2:1b is How to run DSPy with a self-hosted model on a cloud GPU. An agent whose Ollama example sets id to llama3.1:8b is How to run Agno agents with a self-hosted model on a cloud GPU. A TypeScript agent whose local-models section sets url on the model object is How to run Mastra agents with a self-hosted model on a cloud GPU. A ChatAgent whose vLLM example sets ModelPlatformType.VLLM is How to run CAMEL-AI agents 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 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?. The tunnel is How to use SSH port forwarding on a cloud GPU. If the process runs out of memory: How to fix CUDA out of memory on a cloud GPU. If the process raises no kernel image on an RTX 5090: How to fix CUDA no kernel image on an RTX 5090.
The README license line is Apache-2.0.
The commands below are copied from the upstream files named in each step, at the commit in Sources. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked. The README subtitle uses a comparison this page does not copy. A sentence on the README names a closed vendor and a product this page does not copy. The README note names a model this page does not copy. The same note says the original Python project lives on as a community-maintained fork at https://github.com/endolith/open-interpreter. This page did not copy install lines from that fork.
What the docs print about memory
The install page and the models page, read 2026-11-14, do not print a GPU GB figure. The models page’s local-model section says a model trained for a 32K context can still run with Ollama’s smaller server default unless you raise it before starting the server, and it prints OLLAMA_CONTEXT_LENGTH=32768. That line is a context length. It is not vram_gb. The model information table on unsloth/Llama-3.1-8B-Instruct prints parameter counts 8B, 70B, and 405B, and a context length of 128k. This page does not treat those parameter counts as vram_gb. A later section of that card names training hardware. This page does not copy that section and does not use it as an inference memory figure. This page does not mark a catalog row as enough.
GET /v1/public/compute-capabilities does not list GPUs. It returns feature flags. The memory column is vram_gb from GET /v1/public/gpu-catalog (generated_at 2026-11-14T11:11:10.599Z). The catalog response also includes an availability field. This page does not copy it. Every catalog row below has min_gpu_count 1. Rows are sorted by vram_gb, then by name. Which listing you can start is on Deploy.
The catalog template list had no Open Interpreter template. It did have Ollama templates (jupyterlab-ollama, ollama-qwen3, ollama). Those templates are a different install. This page does not treat them as Open Interpreter.
| 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 |
| Ascend 910B | 64 |
| 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
Main of OpenInterpreter/open-interpreter on 2026-11-14 was commit f09c887041133f52da1155e27d29edd8fa1f516e, committed 2026-11-14T01:08:17Z. pushed_at the same day was 2026-11-14T11:01:16Z. The GitHub license field was Apache-2.0. The README and docs/install.md at that commit print this install for macOS and Linux. They do not print a pip install. The CLI reference at that commit does not print --api_base.
curl -fsSL https://www.openinterpreter.com/install | sh
The README says to type i or interpreter to start a session. The install page then prints:
interpreter --version
The quickstart prints this as the open-a-project step. i is the short command for interpreter.
cd my-project
i
The install page says Git is recommended for repository-aware sessions. The same page says user data stays under ~/.openinterpreter, including configuration, sessions, logs, and file-stored credentials. Logs for the interactive TUI are under ~/.openinterpreter/log/. User config is ~/.openinterpreter/config.toml. Project config can live at .openinterpreter/config.toml.
The files this page read do not print an SSH command. The model ports those files do print are in the next section. The tunnel for a port you choose to publish is How to use SSH port forwarding on a cloud GPU.
Local model and an OpenAI-compatible server
The models page prints two built-in local providers:
| Provider | Default base URL | Override |
|---|---|---|
ollama |
http://localhost:11434/v1 |
CODEX_OSS_PORT or CODEX_OSS_BASE_URL |
lmstudio |
http://localhost:1234/v1 |
CODEX_OSS_PORT or CODEX_OSS_BASE_URL |
The launch lines on that page are:
interpreter --oss --local-provider ollama
interpreter --oss --local-provider lmstudio
--oss without --local-provider uses the saved oss_provider or opens a picker. The same page says not to create a separate model_providers entry just to change the address of either built-in local provider. The override example it prints is a different host from the default table:
CODEX_OSS_BASE_URL=http://192.168.1.20:1234/v1 \
interpreter --oss --local-provider lmstudio -m qwen/qwen3-coder-next
This page does not merge http://localhost:11434/v1 and http://192.168.1.20:1234/v1 into one URL. The CLI reference prints --model, -m as the model override. It does not print --api_base.
For Qwen models on Ollama, the models page says the qwen-code harness uses the Chat Completions transport, and it prints:
interpreter exec --oss --local-provider ollama --chat-completions \
-m qwen2.5-coder:7b "inspect this project"
The providers page says --chat-completions selects that request shape for one invocation and does not change the provider’s base URL, credentials, or model. The same page’s custom provider block, which is not the local-provider command, is:
[model_providers.example]
name = "Example"
base_url = "https://api.example.com/v1"
env_key = "EXAMPLE_API_KEY"
wire_api = "chat"
The configuration page’s custom provider example sets model_provider to acme, model to acme-coder-large, base_url to https://api.acme.example/v1, and wire_api to responses. Those two wire_api values are not the same. This page does not merge the two base URLs. The configuration page also prints a hosted gateway example this page does not copy. The common-settings example on that page names a model and a provider this page does not copy.
The README prints a /harness list. This page copies the entries native, qwen-code, swe-agent, and minimal. The other entries name products this page does not copy.
The Open Interpreter files this page read do not print hf download. The block below is added on this page. How to download a Hugging Face model prints pip install -U "huggingface_hub" and hf download. This page did not download the repository. The Open Interpreter files this page read do not print this repository id. qwen2.5-coder:7b and qwen/qwen3-coder-next are the strings the models page prints. This page does not treat them as this repository id.
pip install -U "huggingface_hub"
hf download unsloth/Llama-3.1-8B-Instruct
On 2026-11-14, GET https://huggingface.co/api/models/unsloth/Llama-3.1-8B-Instruct reported gated false, cardData.license llama3.1, pipeline_tag text-generation, lastModified 2025-05-09T19:32:57.000Z, and sha 4699cc75b550f9c6f3173fb80f4703b62d946aa5. The card YAML also includes extra_gated_prompt. Those two fields are not the same fact. The card’s base_model field names a repository this page does not copy. This page did not run hf download.
Session and safety
The quickstart says the first run walks through provider setup, and that /model changes the provider later. A request it prints is add a /health endpoint that returns the build sha. It says Open Interpreter reads the project, proposes work, edits files, and runs commands through the active sandbox and approval settings. Resume is interpreter resume --last. A one-shot command the same page prints is interpreter exec "summarize the current diff".
The sandbox page says there are two controls. Sandbox mode is the boundary for local command execution. Approval policy is when the agent pauses and asks. /permissions inspects or changes the active posture. The modes it prints are read-only, workspace-write, and danger-full-access. The policies it prints are untrusted, on-request, and never. The defaults it prints are:
sandbox_mode = "workspace-write"
approval_policy = "on-request"
The same page says --yolo and --dangerously-bypass-approvals-and-sandbox remove both approval prompts and sandboxing, and to use them only inside an external sandbox such as a throwaway VM or isolated container. The cyber-safety page says to keep security work scoped to assets you own or are authorized to test, and it says not to ask the agent to exploit third-party systems.
The streaming web chat page says to run node bridge.mjs and open http://127.0.0.1:8787/. It says Node 20 or newer is needed. It says the bridge is loopback-only and says not to forward that port. This page does not print the two source files from that page and does not treat port 8787 as the model API.
When it fails
CUDA out of memory. The Open Interpreter files this page read print no GPU GB figure. 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 Open Interpreter README and the docs pages this page read do not name compute capability 12.0. 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 this program on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only.
Model metadata for ... not found. The models page says this means the local server returned a model ID that is not in Open Interpreter’s compatibility catalog. It says that message does not by itself mean the server connection failed. It says to confirm the exact ID the server exposes and pass that same value with -m.
Truncated context. The models page says that if the server truncates the beginning of the request, a model may answer from guesses, print a tool call as ordinary text, or ask you to run a command instead of invoking a tool. It says those are model or server compatibility symptoms. The context-length command it prints is OLLAMA_CONTEXT_LENGTH=32768 ollama serve.
What to copy off, then destroy
The install page says the managed uninstall keeps user data under ~/.openinterpreter. It says deleting that directory also erases local configuration, sessions, logs, and file-stored credentials, and that the deletion cannot be undone. It says that deletion does not remove credentials kept in an OS keyring or environment variables. Copy ~/.openinterpreter, including config.toml and log/, and copy the project files the session edited, 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 billing page 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.
Amounts are not copied here.
Frequently asked questions
What memory figure do the Open Interpreter docs print?
The install page and the models page do not print a GPU GB figure. The models page’s local-model example says a model trained for a 32K context can still run with Ollama’s smaller server default, and it prints OLLAMA_CONTEXT_LENGTH=32768. That is a context length. It is not vram_gb. The unsloth/Llama-3.1-8B-Instruct model information table prints parameter counts 8B, 70B, and 405B, and context length 128k. This page does not treat those figures as vram_gb. This page does not mark a catalog row as enough. Which listing you can start is on https://cloudgpu.app/deploy.
Which install lines do the docs print?
The README and the install page print curl -fsSL https://www.openinterpreter.com/install | sh for macOS and Linux, then interpreter or i to start a session. They do not print a pip install. Main on 2026-11-14 was commit f09c887041133f52da1155e27d29edd8fa1f516e. This page did not run the command.
How do the docs point the program at a server on the same machine?
The models page sets the ollama provider’s default base URL to http://localhost:11434/v1 and the launch command to interpreter --oss --local-provider ollama. An override example sets CODEX_OSS_BASE_URL to http://192.168.1.20:1234/v1 with --local-provider lmstudio. A custom provider block sets base_url and wire_api to chat. The CLI reference prints --model and does not print --api_base. The files do not print an SSH command. The streaming web chat page listens on http://127.0.0.1:8787/ and says not to forward that port.
What license sentence does the README print?
The README license line is Apache-2.0. The GitHub license field on 2026-11-14 was Apache-2.0. That license text is not a CloudGPU rate.
Sources (read 2026-11-14, Asia/Shanghai)
- OpenInterpreter/open-interpreter README at commit
f09c887041133f52da1155e27d29edd8fa1f516e, committed 2026-11-14T01:08:17Z. GitHub license field Apache-2.0.pushed_at2026-11-14T11:01:16Z. Install commandcurl -fsSL https://www.openinterpreter.com/install | sh. Session commandsiandinterpreter. License line Apache-2.0. The subtitle comparison, a closed-vendor sentence, and a model name on that page are not copied. The note points at https://github.com/endolith/open-interpreter. This page did not copy install lines from that fork docs/install.mdat that commit.interpreter --version. User data under~/.openinterpreter. Logs under~/.openinterpreter/log/. No pip install. No GPU GB figuredocs/quickstart.mdat that commit.cd my-projecttheni./model.interpreter resume --last.interpreter exec "summarize the current diff". A sentence on that page names a sign-in product this page does not copydocs/models.mdat that commit. Ollama defaulthttp://localhost:11434/v1. LM Studio defaulthttp://localhost:1234/v1. OverrideCODEX_OSS_BASE_URLexamplehttp://192.168.1.20:1234/v1with-m qwen/qwen3-coder-next.OLLAMA_CONTEXT_LENGTH=32768.interpreter exec --oss --local-provider ollama --chat-completions -m qwen2.5-coder:7b.Model metadata for ... not found. No--api_base. No GPU GB figuredocs/cli-reference.mdat that commit.--model, -m.--oss.--local-providervaluesollamaandlmstudio. No--api_basedocs/providers.mdat that commit.wire_api = "chat"for OpenAI-compatible Chat Completions. Examplebase_urlhttps://api.example.com/v1.--chat-completionsdoes not change the base URL. A hosted gateway example on that page is not copieddocs/config.mdat that commit. User config~/.openinterpreter/config.toml. Acme examplebase_urlhttps://api.acme.example/v1andwire_apiresponses. The common-settings model id is not copieddocs/sandbox.mdanddocs/cyber-safety.mdat that commit.sandbox_mode = "workspace-write".approval_policy = "on-request".--yoloremoves approvals and sandboxing. Cyber-safety says not to ask the agent to exploit third-party systemsdocs/streaming-web-chat.mdat that commit.http://127.0.0.1:8787/. Node 20 or newer. The page says not to forward that port. The two source files are not copieddocs/open-interpreter-delta.mdat that commit.--chat-completionsselectswire_api = "chat". That page is a compatibility checklist. Product names on it are not copied- Hugging Face model API
unsloth/Llama-3.1-8B-Instruct:gatedfalse,cardData.licensellama3.1,pipeline_tagtext-generation,lastModified2025-05-09T19:32:57.000Z,sha4699cc75b550f9c6f3173fb80f4703b62d946aa5. The card YAML includesextra_gated_prompt. The model information table prints 8B, 70B, 405B, and 128k.base_modelnames a repository this page does not copy.hf downloadis added on this page and was not run. The training-hardware section is not copied - Public GPU memory: https://cloudgpu.app/v1/public/gpu-catalog (
generated_at2026-11-14T11:11:10.599Z).GET /v1/public/compute-capabilitiesreturned feature flags and no GPU list. The catalog had no Open Interpreter template
Cards you can start are on the Deploy page. Amounts are not copied here.
Last updated: 2026-11-14 (Asia/Shanghai).