Free tools Windows power users keep installed
One-click scans. No signup required.
A hybrid AI workflow can keep selected prompts and files on your computer while sending harder tasks to Claude—but only when you deliberately route those tasks to a local model. Claude Code runs on your machine, yet Anthropic says it sends the portions of files needed for the current task to its API. The distinction is the endpoint that handles inference, not simply where the coding tool is installed.
What “local” means in a hybrid setup
There are three separate questions to answer before using this workflow:
As an Amazon Associate I earn from qualifying purchases.
- Where does the tool run? Claude Code is installed and runs on your computer.
- Where does inference happen? A request configured for Ollama goes to a local model; a request sent to Claude goes to Anthropic’s API.
- What terms govern the data? Retention and model-training policies depend on the Claude product, account, and applicable organization terms.
Anthropic’s Claude Code FAQ says the tool reads source files locally and sends only the portions needed for the current task to its API. That means a Claude Code session is not wholly on-device just because the application runs locally. Anthropic also states in its setup documentation: “Network: Internet connection required for authentication and AI processing.”
How to route work between Ollama and Claude
Ollama documents an Anthropic Messages API-compatible connection for Claude Code, including a quick launch and a manual configuration. With this arrangement, Claude Code can connect to a selected Ollama model through a local endpoint; when you want Claude’s cloud inference, use the Claude connection and treat that request as cloud processing.
#1 Best Overall
Quick setup
- Install and configure Ollama and Claude Code according to their current official setup instructions.
- Run
ollama launch claudeto use Ollama’s documented quick setup for launching Claude Code with Ollama. - Choose an Ollama model supported by the current integration, then test with a task that does not contain sensitive material. Confirm that the intended endpoint is being used before sharing project files or prompts.
Manual setup
Ollama’s manual example configures the Anthropic-compatible endpoint with these environment variables, then starts Claude Code with a selected Ollama model:
export ANTHROPIC_AUTH_TOKEN=ollama
export ANTHROPIC_BASE_URL=http://localhost:11434
claude --model qwen3-coderThe exact model names and setup behavior can change. Check Ollama’s current Anthropic API compatibility instructions for the latest supported commands and model recommendations; its documentation names qwen3-coder and glm-4.7 among coding options.
A practical routing rule
- Use the local endpoint for work you want to keep on-device, provided the model can handle it and your machine has the required capacity.
- Route a difficult task to Claude only when you are willing to send the necessary prompt and file excerpts to Anthropic for cloud processing.
- Before either route, review what the coding agent can read and include. Limit access to irrelevant files and avoid placing credentials or secrets in prompts.
What changes when you route a task
| Factor | Ollama local model | Claude via Anthropic |
|---|---|---|
| Content destination | Requests configured to Ollama’s local endpoint are processed by the local model. | Claude Code sends the portions of files needed for the task to Anthropic’s API, according to the Claude Code FAQ. |
| Internet | Local inference can run against the local endpoint; the exact workflow’s other requirements depend on its setup. | Anthropic says Claude Code requires internet for authentication and AI processing in its setup documentation. |
| Hardware and context | Depends on the chosen model and context length. Ollama says its 30B-parameter Qwen 3 coder example needs at least 24 GB of VRAM to run smoothly, and longer context lengths require more. | Runs through Anthropic’s cloud service; the cited materials do not establish a directly comparable local hardware requirement. |
| Model suitability | Depends on whether the selected local model can do the task within available hardware and context capacity. | Useful when you choose Claude’s cloud inference for a task, with the associated data transfer and applicable account terms. |
The 24 GB figure is specific to Ollama’s Qwen 3 coder example, not a minimum for all local AI. It is not a general recommendation to buy a particular graphics card. No controlled head-to-head quality or speed result is established by these sources, so the workflow should not be treated as proof that either route is universally faster or better.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Check the privacy terms for the account you use
Sending a task to Claude is a separate decision from installing Claude Code. Anthropic’s Privacy Center article covers consumer Free, Pro, and Max accounts, including their Claude Code use, and says chats and coding sessions may be used to improve models in specified situations, including user opt-in and safety review. Read the current model-training policy for the applicable consumer account rather than assuming one rule applies to every Claude product.
Rank #3
Commercial products and API usage have separate terms and retention arrangements. Anthropic’s API and data retention documentation describes feature-specific arrangements. If you use Claude through an organization or API integration, verify the terms that actually govern that account and feature; do not infer them from the consumer policy.
Quick Recap
Choose the workflow by the data and task
- Keep the task local when its contents should not be sent to a cloud inference endpoint and the local model is adequate for the job.
- Use Claude deliberately when you want its cloud inference for a harder task and have checked that sharing the relevant project excerpts is acceptable under your account’s terms.
- Recheck endpoint and access whenever you change configuration, model, or project permissions. A familiar local interface does not by itself confirm where inference is happening.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
