Ollama

Run open models locally or in the cloud

by Ollama Inc.

Overview

Ollama is a runtime for open-weight language models that runs on your own machine. It is installed from ollama.com for macOS, Windows or Linux and used either as a desktop app or from a terminal, downloads a model such as gemma4 or qwen3.5, and serves it on http://localhost:11434 through Ollama's own API and through OpenAI-, Anthropic- and Responses-shaped endpoints so another program can talk to it (ollama.com/pricing and docs.ollama.com/quickstart, read 12 October 2026). The documentation also covers pointing a coding agent at that local server: "ollama launch claude" and the manual route that exports ANTHROPIC_BASE_URL=http://localhost:11434 are both written up (docs.ollama.com/integrations/claude-code). It is sold freemium. Free is $0 and lists "Run models locally" as included; Pro is $20 / mo. or $200 per year billed annually, for which the same page prints the annual equivalent of $16.67 / mo. billed annually, and Max is $100, Team $500 and Enterprise quoted. The paid tiers buy usage credits for cloud models rather than the local runtime - the FAQ states that running models on your own hardware is always unlimited, and cloud usage is measured in tokens - and the page prints no currency code and names no region. The one eligibility line in the terms is age: you must be at least 18 years old to use the services.

The split between the local and the cloud half is documented on both sides. The privacy policy states that Ollama runs on your local device and that it does not collect, store, transmit or have access to your prompts, responses, model interactions or other content you process locally, with your data staying on your machine; it adds that cloud-hosted models are processed transiently, that inputs and outputs are not used to train models, and that limited device and usage metadata excluding prompt and response content is collected. The terms repeat the training position and add that you retain ownership of what you create and that Ollama claims no ownership over inputs or outputs. Two consequences follow from those pages and not from any test of ours: a cloud model is a different route from a local one and sends the prompt to Ollama's infrastructure, and the local route still needs an editing surface, which Ollama documents as a separate agent rather than supplying itself.

On the three boundaries this catalogue tracks, two can be sourced here and one cannot. Training and data use is stated without qualification: "We do not use your inputs or outputs to train AI models." Commercial use is not established: the terms describe a limited licence to process your content in order to provide the service, and they neither grant permission to use the service for commercial purposes nor prohibit it. Whether an API is included in a consumer plan has a source, and for the local half the answer is yes, since the local server is documented without an API key, while cloud models run on a key against a usage-credit balance at the published per-million-token rates - so the two halves are measured differently.

Editor's take

Reach for it when the deciding requirement is that the model runs on your own machine and the runtime has to cost $0, and read it as a foundation rather than a finished coding setup: Ollama supplies the runtime, the local endpoint and a command line, and its documentation shows how to point an existing agent at that endpoint. Two things to check before treating it as a route for private work. The same pricing page sells cloud models, so the model you pick decides whether prompts leave the machine. And no page read here states that a third-party agent pointed at localhost keeps everything else it does on the device, so the editing half is the part left to verify.

Our reading of who this suits, based on the sources above. Opinion, offered to help you decide - not a lab result.

Key Features

•Runs open-weight models on macOS, Windows and Linux, as a desktop app or from a terminal
•Free plan listed as including running models locally
•Local server on http://localhost:11434, documented without an API key
•Documented setups for using local models from Claude Code, Codex CLI and OpenCode
•Cloud models bought separately on usage credits priced per million tokens
•Command line for pulling, serving and chatting with models

Use Cases

→Chatting with a model on a laptop that is not connected to the internet
→Serving a local endpoint for an editor or agent that has no model of its own
→Comparing open models on your own hardware before committing to a hosted API
→Drafting against a private repository without sending it to a hosted service

Pros

✓The privacy policy states that content processed locally is not collected, transmitted or accessible to Ollama
✓Running models on your own hardware is documented as always unlimited, and the local runtime is $0
✓The training position is stated plainly: inputs and outputs are not used to train AI models
✓The local endpoint is documented at localhost, so another editor or agent can be pointed at it

Cons

✕It is a model runtime rather than an editor, so the repository-editing half has to come from another tool
✕Cloud models sit on the same pricing page and send prompts to Ollama, so the model choice sets the data path
✕No page read here states that an agent pointed at the local endpoint keeps its other traffic on the device
✕The terms do not grant commercial use of the service, and the prices are printed without a currency code or a region

Frequently Asked Questions

Is Ollama free?

The Free plan is $0 and the pricing page lists running models locally as included; Pro is $20 a month or $200 billed annually, Max is $100 a month, Team starts at $500 and Enterprise is quoted. The paid tiers buy usage credits for cloud models, and the page states that running models on your own hardware is always unlimited. The figures are printed with a dollar sign only, with no currency code and no region.

Do my prompts leave my machine?

For local models the privacy policy says no: Ollama runs on your local device, does not collect, store, transmit or have access to prompts, responses or other content processed locally, and your data stays on your machine. Cloud-hosted models are the other route on the same product; those requests go to Ollama and are processed transiently, and the page states that prompt or response data is never logged or trained on. The model you choose therefore decides the data path.

Can I use it on a private codebase?

The local half is documented as staying on the device, and the documentation shows how to point Claude Code, Codex CLI and OpenCode at http://localhost:11434 so the model runs locally. What no page read here states is that everything else such an agent does also stays local, and we did not capture traffic to settle it, so treat the editing surface as the part still to verify.

Does Ollama train on what I run through it?

The terms state that Ollama does not use your inputs or outputs to train AI models, and the privacy policy repeats that prompt and response content is not trained on; the metadata it does collect is described as device and usage information that excludes prompt and response content. Cloud requests are processed transiently and are not stored beyond what fulfilling the request needs.

Task guides for this product

T7 · Check whether code and prompts stay on your Mac

Key facts and where they come from

PriceNot established: The live page lists Pro $20/month or $200/year, Max $100/month and separate cloud usage credits/token rates; local running remains separate. ISO currency code and applicable purchase-region/checkout conditions remain unestablished.Source: Ollama - pricing
AvailabilityDocumented: 'Ollama downloads the model and starts a chat on your computer.' / 'ollama run gemma4:e2b'. The local server answers without a key: 'curl http://localhost:11434/api/chat', with OpenAI-, Anthropic- and Responses-compatible paths on the same host.Source: Ollama - quickstart
Commercial useNot established: Ollama preserves user ownership and requires lawful use, but does not grant a general commercial-output licence overriding the chosen local/cloud model’s own licence and third-party rights. Those actual-model conditions remain unverified.Source: Ollama - terms of service · Ollama - quickstart
Training on your dataDocumented: 'Last updated: March 2026'. 'Ollama runs on your local device. We do not collect, store, transmit, or have access to your prompts, responses, model interactions, or other content you process locally. Your data stays on your machine.' / 'Ollama runs locally. We don't see your prompts or data when you run locally.' / 'When using cloud-hosted models, we process this content transiently to provide the Service and this content is not stored beyond the time required to fulfill the request.' / 'We do not use your inputs or outputs to train any AI models or request prompt or response content in support requests.' Limited metadata 'that does not include your prompt or response content' is collected. The policy is by 'Ollama Inc.'Source: Ollama - privacy policy
API billingDocumented: 'Download Ollama for macOS, Windows, or Linux.' / 'Use cloud models with an API key, or run models locally without one.' / 'Ollama downloads the model and starts a chat on your computer.' / 'ollama run gemma4:e2b'. The local server answers without a key: 'curl http://localhost:11434/api/chat', with OpenAI-, Anthropic- and Responses-compatible paths on the same host. The page also documents launching agents: 'Use a coding agent … ollama launch claude' and 'ollama launch codex'. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Cloud models are priced per million input, cached and output tokens, with off-peak rates on weekdays and weekends.Source: Ollama - quickstart · Ollama - pricing
Export formatsNot established: The quickstart establishes local/cloud model execution and APIs, but not downloadable conversation/report file formats or surface/plan export boundaries. API response encoding is not a finished-report export format.Source: Ollama - quickstart
Usage limitsDocumented: Running models on local hardware remains unlimited in the pricing-page wording; paid subscriptions provide cloud usage credits, not a limit on the local runtime. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Free includes 1 concurrent request, Pro 3, and Max and Team 10.Source: Ollama - pricing

Official sources

The vendor pages we read for the claims on this page, what each one states, and the day we read it.

Ollama - privacy policy ↗
'Last updated: March 2026'. 'Ollama runs on your local device. We do not collect, store, transmit, or have access to your prompts, responses, model interactions, or other content you process locally. Your data stays on your machine.' / 'Ollama runs locally. We don't see your prompts or data when you run locally.' / 'When using cloud-hosted models, we process this content transiently to provide the Service and this content is not stored beyond the time required to fulfill the request.' / 'We do not use your inputs or outputs to train any AI models or request prompt or response content in support requests.' Limited metadata 'that does not include your prompt or response content' is collected. The policy is by 'Ollama Inc.'
Read on October 12, 2026
Ollama - pricing ↗
The body lists Pro $20/month or $200/year and Max $100/month plus cloud token/credit billing and local unlimited running, but no ISO currency code or purchase-region guarantee. Hosting in US/Europe is not a sales-country list. Historical registered scope read on 2026-10-15, not rechecked in full on 2026-10-15: Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: Free '$0' with 'Use any cloud model with pay-as-you-go usage credits.' and 'Includes: Run models locally'; Pro '$20 / mo. or $200/yr billed annually'; the annual toggle prints '$16.67 / mo. billed annually'; Max '$100 / mo.'; Team '$500 / mo.'; Enterprise 'Custom'. FAQ: 'Running models on your own hardware is always unlimited.' / 'Free includes 1 concurrent request, Pro 3, and Max and Team 10.' / 'Where are models hosted? Ollama hosts models and compute resources primarily in the United States.' / 'Is my prompt or response data trained on? Prompt or response data is never logged or trained on.' Cloud models are priced per million input, cached and output tokens, with off-peak rates on weekdays and weekends. No currency code and no region are stated for the plan prices. Official body read on 2026-10-15, limited to this scope: Free includes running models locally; paid plans include cloud usage credits and local hardware running remains unlimited by the vendor wording. The table prints$without a currency code/market; subscription credit pools are not the local runtime. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Cloud models are priced per million input, cached and output tokens, with off-peak rates on weekdays and weekends. Running models on local hardware remains unlimited in the pricing-page wording; paid subscriptions provide cloud usage credits, not a limit on the local runtime. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: Free includes 1 concurrent request, Pro 3, and Max and Team 10.
Read on October 15, 2026
Ollama - terms of service ↗
The terms preserve user ownership and impose lawful use/third-party rights conditions; they do not grant an unconditional commercial-output licence or override each chosen model licence. Historical registered scope read on 2026-10-15, not rechecked in full on 2026-10-15: Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: 'Last updated: May 2026', from 'Ollama Inc.'. Section 2: 'You must be at least 18 years old to use our services.' Section 5: 'We do not use your inputs or outputs to train AI models.' Section 7: 'You retain ownership of content you create using our services. By using our services, you grant us a limited, non-exclusive license to process your content solely to provide the Service. We claim no ownership over your inputs or outputs.' Section 6: payments are processed by Stripe, subscriptions renew automatically unless cancelled, and 'Purchased usage credits expire one year after they are added to your account'. The restrictions prohibit competing-product development, reverse engineering and unpermitted automated access. The page grants no commercial-use permission and contains no clause forbidding it. Official body read on 2026-10-15, limited to this scope: Terms claim no ownership of user inputs/outputs and no model training on them; the text neither grants nor forbids commercial use. Purchased usage credits expire one year after addition. The displayed price’s missing currency/region is a separate pricing-page observation.
Read on October 15, 2026
Ollama - quickstart ↗
The quickstart documents local/cloud model running and APIs; it does not specify downloadable conversation/report file formats or their surface/plan boundaries. Historical registered scope read on 2026-10-12, not rechecked in full on 2026-10-15: 'Download Ollama for macOS, Windows, or Linux.' / 'Use cloud models with an API key, or run models locally without one.' / 'Ollama downloads the model and starts a chat on your computer.' / 'ollama run gemma4:e2b'. The local server answers without a key: 'curl http://localhost:11434/api/chat', with OpenAI-, Anthropic- and Responses-compatible paths on the same host. The page also documents launching agents: 'Use a coding agent … ollama launch claude' and 'ollama launch codex'.
Read on October 15, 2026
Ollama - Claude Code integration ↗
'Claude Code reads your codebase, edits files, and runs commands. Ollama connects Claude Code to local and cloud models through its Anthropic-compatible API.' Launch with 'ollama launch claude', or set it up by hand: 'export ANTHROPIC_BASE_URL=http://localhost:11434' then 'claude --model qwen3.5'. The cloud route is documented separately and points at https://ollama.com. The page does not state that the agent's traffic stays on the device.
Read on October 12, 2026

Compare with other tools in Code Tools

Product	Maker	Pricing	Launched	Shared tags
LM Studio	Element Labs, Inc.	Freemium	Launch date not verified	On-device, Open weights
Claude Code	Anthropic	Paid	2025-02-24	Terminal
MonkeyCode	Chaitin Tech	Freemium	2025-06	—
Lovable	Lovable	Freemium	2024-12	—
Devin Desktop	Cognition	Freemium	2024-11-13	—
Replit Agent	Replit	Freemium	2024-09	—