Apps, command-line tools and editors on your machine
| Product | Models from different providers in one sessionOne session can use models from different providers, for example one agent on one maker's model and another on a second maker's. Switching the single model for the whole session does not count on its own. A tool that is itself a gateway or library routing to the providers its page names counts. | Several endpoints, built in (no proxy)The tool itself can hold more than one provider or address, so nothing has to sit in front of it to reach different providers. | A model on your machineThe tool can use a model running on your own computer, for example through Ollama, LM Studio, oMLX or llama.cpp. | A model in a container environment you runA model server you host in containers, on your own server or in your own cloud account. A model in a cloud provider's model service, such as Amazon Bedrock, is a cloud model, not one you run. | Local and cloud in one sessionA local model (on your machine or in a container environment you run) and a cloud model (in its model company's cloud or a cloud provider's model service) can both be used in the same session or setup. |
|---|
| Aider | | Partly | YesHow to set it up | | |
| Claude Code | No | | Via a proxy | | Via a proxy |
| Claude Desktop | | Partly | | | |
| Cline | PartlyHow to set it up | | | | |
| Codex | No | YesHow to set it up | | | PartlyHow to set it up |
| Continue | PartlyHow to set it up | YesHow to set it up | YesHow to set it up | | YesHow to set it up |
| Cursor | PartlyHow to set it up | PartlyHow to set it up | Not found | | |
| Factory (Droid CLI) | PartlyHow to set it up | YesHow to set it up | YesHow to set it up | | PartlyHow to set it up |
| Gemini CLI | PartlyHow to set it up | Not found | Partly | | PartlyHow to set it up |
| GitHub Copilot | PartlyHow to set it up | PartlyHow to set it up | YesHow to set it up | | YesHow to set it up |
| goose | YesHow to set it up | YesHow to set it up | YesHow to set it up | Yes | YesHow to set it up |
| Kilo Code | YesHow to set it up | PartlyHow to set it up | YesHow to set it up | | PartlyHow to set it up |
| OpenCode | PartlyHow to set it up | YesHow to set it up | YesHow to set it up | | YesHow to set it up |
| Perplexity app (Personal Computer and Portable Computer) | YesHow to set it up | Not found | YesHow to set it up | | YesHow to set it up |
| promptfoo | YesHow to set it up | YesHow to set it up | YesHow to set it up | | YesHow to set it up |
| Vellum (assistant) | PartlyHow to set it up | | Claimed | | PartlyHow to set it up |
| Zed | PartlyHow to set it up | | YesHow to set it up | | YesHow to set it up |
Words in the cells: YesA page we read says so., PartlyA page says part of it, for example separate models per agent without saying the providers can differ., Via a proxyOnly with a proxy, or a local server that imitates a cloud provider's interface, in front of the tool., NoA page we read says the product does not support it., ClaimedA vendor page says so on a marketing or overview page, and we found no guide showing how to set it up., Not foundWe read pages for it and found nothing there. It says nothing about whether the product has the feature., n/aThe question does not apply to the product, for example because it is a model server and not a tool that connects to models., BlankNothing is published for it yet.
Libraries that run in your own code
| Product | Models from different providers in one sessionOne session can use models from different providers, for example one agent on one maker's model and another on a second maker's. Switching the single model for the whole session does not count on its own. A tool that is itself a gateway or library routing to the providers its page names counts. | Several endpoints, built in (no proxy)The tool itself can hold more than one provider or address, so nothing has to sit in front of it to reach different providers. | A model on your machineThe tool can use a model running on your own computer, for example through Ollama, LM Studio, oMLX or llama.cpp. | A model in a container environment you runA model server you host in containers, on your own server or in your own cloud account. A model in a cloud provider's model service, such as Amazon Bedrock, is a cloud model, not one you run. | Local and cloud in one sessionA local model (on your machine or in a container environment you run) and a cloud model (in its model company's cloud or a cloud provider's model service) can both be used in the same session or setup. |
|---|
| AG2 | PartlyHow to set it up | YesHow to set it up | YesHow to set it up | | PartlyHow to set it up |
| LiteLLM | YesHow to set it up | YesHow to set it up | YesHow to set it up | | YesHow to set it up |
| LlamaIndex | | YesHow to set it up | YesHow to set it up | | PartlyHow to set it up |
| Pydantic AI | YesHow to set it up | YesHow to set it up | YesHow to set it up | | PartlyHow to set it up |
Products that run in more than one place
These come in forms that run in more than one place, so each is on every page that fits one of its forms.
On this page as command-line tool, editor extension and desktop app. Also as hosted or cloud agent, on the page Hybrid as a cloud service.
On this page as command-line tool, editor extension and desktop app. Also as hosted or cloud agent, on the page Hybrid as a cloud service.
Tasks run in Codex Cloud or on the machine through the local clients (ChatGPT desktop app, CLI, IDE extension); where a task runs is chosen per task.
- Conditions: which plan allows cloud tasks not read
On this page as editor extension, command-line tool and desktop app. Also as hosted or cloud agent, on the page Hybrid as a cloud service.
On this page as editor extension and command-line tool. Also as hosted or cloud agent, on the page Hybrid as a cloud service.
On this page as desktop app. Also as hosted or cloud agent, on the page Hybrid as a cloud service.
On Mac, the app, files, app control and the local model run on the machine; planning, web search, reasoning and cloud models run in Perplexity's cloud. On Windows and Linux with supported hardware, the model and agent run on the machine and single steps go to the cloud only with approval.
- Hardware: Mac: Apple silicon, macOS 15 or later, 24 GB unified memory minimum (the page says 32 GB gives better results). Windows and Linux (Portable Computer): an NVIDIA RTX or RTX PRO card with at least 24 GB of video memory, or a DGX Spark; on Windows, per the Personal Computer page only (see By page), an AMD Ryzen AI Max processor with at least 24 GB of GPU-accessible memory
- Plan: Local models: Pro, Max or Enterprise. Portable Computer: Pro or Max, personal or Enterprise
- Scope: Hybrid compute is Mac-only; with Portable Computer, Windows and Linux machines run the task on the device and send single steps to the cloud only with the user's approval
- By page: The Portable Computer page (read 2026-10-22) names only NVIDIA hardware and offers Qwen 27B on supported Windows and Linux PCs; the Personal Computer page (read 2026-10-22) adds AMD Ryzen AI Max on Windows and limits Qwen 27B to AMD and DGX Spark.
On this page as desktop app. Also as hosted or cloud agent, on the page Hybrid as a cloud service.
By default the assistant runs locally on a Mac; Vellum Cloud runs it in an isolated container.
- Conditions: Mac for the local form
Where their models run
Where each product's models run, as its vendor's pages state it.
| Product | Models runWhere the vendor's own pages say the product's models run, not a limit: your machine, a container environment you run, the model company's cloud (servers of the company that makes or serves the model), or a cloud provider's model service. For a cloud, read the data and compliance lines of the company running it, linked by its name. |
|---|
| AG2 | Your machine |
| Aider | Your machine and the model company's cloud (OpenAI, Anthropic) |
| Claude Desktop | Your machine, the model company's cloud (Anthropic, Ollama) and a cloud provider's model service (Bedrock, Vertex, Foundry) |
| Cline | Your machine and a cloud provider's model service (Azure, GCP Vertex) |
| Codex | A cloud provider's model service (Amazon Bedrock) |
| Continue | Your machine, the model company's cloud (Gemini) and a cloud provider's model service (Amazon Bedrock) |
| Cursor | The model company's cloud (OpenAI, Anthropic) and a cloud provider's model service (AWS Bedrock, Azure) |
| Factory (Droid CLI) | Your machine and the model company's cloud (OpenAI, Anthropic) |
| Gemini CLI | The model company's cloud (Gemini API) and a cloud provider's model service (Vertex AI) |
| GitHub Copilot | Your machine and the model company's cloud (Anthropic, OpenAI) |
| goose | Your machine, a container environment you run and a cloud provider's model service (GCP Vertex AI) |
| Kilo Code | Your machine and the model company's cloud |
| LiteLLM | Your machine, the model company's cloud (OpenAI, Anthropic, Gemini) and a cloud provider's model service (Bedrock, Azure) |
| LlamaIndex | Your machine |
| OpenCode | Your machine and the model company's cloud (OpenAI) |
| Perplexity app (Personal Computer and Portable Computer) | Your machine and the model company's cloud (Perplexity) |
| promptfoo | Your machine, the model company's cloud (OpenAI, Anthropic) and a cloud provider's model service (Azure, Bedrock) |
| Pydantic AI | Your machine, the model company's cloud and a cloud provider's model service |
| Vellum (assistant) | Your machine and the model company's cloud (OpenAI, Anthropic, Gemini) |
| Zed | Your machine |
Our summary of the vendor's pages as of the dates shown on each line (2026-09-24 to 2026-10-23); confirm on the vendor's page.
Rendered 2026-10-28
The local piece: model servers
A model server loads a model onto your machine and answers requests from the tools above. The table shows what each server's pages say about the two formats a tool can speak to it, and whether its pages cover Claude Code.
| Server | Anthropic-format endpoint | OpenAI-format endpoint | Claude Code page |
|---|
| Ollama | Yes (docs.ollama.com page, read 2026-10-21) | Yes (docs.ollama.com page, read 2026-10-19) | Yes (docs.ollama.com page, read 2026-10-21) |
| LM Studio | Yes (lmstudio.ai page, read 2026-10-21) | Yes (lmstudio.ai page, read 2026-10-20) | Yes (lmstudio.ai page, read 2026-10-21) |
| llama.cpp server | Yes (github.com page, read 2026-10-21) | | Not found in the one page read (the server README) (github.com page, read 2026-10-21) |
| oMLX | Yes (github.com page, read 2026-10-21) | Yes (github.com page, read 2026-10-21) | Not found in the pages we read (README only; it mentions Claude Code in a context-window and keep-alive note, with no setup steps) (github.com page, read 2026-10-21) |
| MLX-LM | Not found in the pages we read (github.com page, read 2026-10-21) | Partly (github.com page, read 2026-10-21) | Not found in the one page read (the server README) (github.com page, read 2026-10-21) |
Servers that run in more than one place
Model servers we have not read for these questions: Atomic-Chat, BentoML.
Claude Code with a local server
This section is about pointing Claude Code at a model server that runs on your own machine.
Anthropic's statement
- The Claude Code gateway page says Anthropic provides no support when a gateway sends Claude Code's requests to models that are not Claude, whichever gateway it is, and that Anthropic neither endorses, maintains nor audits gateway products made by others. (code.claude.com page, read 2026-10-21)
The pages do not say whether this statement covers a local model server, and we do not say either.
The setting Claude Code uses
- The gateway page names ANTHROPIC_BASE_URL as the setting that sends Claude Code's requests to a gateway. (code.claude.com page, read 2026-10-21)
What each page says
- Ollama links Claude Code with local and cloud models using its Anthropic-compatible API. (docs.ollama.com page, read 2026-10-21)
- LM Studio's blog post says version 0.4.1 introduced an endpoint at /v1/messages that follows Anthropic's format, and that local models can be used with Claude Code through it. (lmstudio.ai page, read 2026-10-21)
- Under its POST /v1/messages heading, the llama.cpp server README says compatibility with Anthropic's API specification is not strongly claimed, and that in the authors' experience the endpoint is enough for many apps. (github.com page, read 2026-10-21)
- The oMLX README lists a POST /v1/messages endpoint described as the Anthropic Messages API, next to OpenAI-style endpoints on the same server. (github.com page, read 2026-10-21)
- The MLX-LM server document says its HTTP API is meant to resemble OpenAI's chat API. An Anthropic Messages endpoint and Claude Code are not found in that document, the only page read. (github.com page, read 2026-10-21)