Ollama
Run a small language model on your own computer and complete one local conversation.
- Suitable for
- People who want to see what local AI means before connecting other apps to it.
- Do not start here if
- Not suitable for old computers with little free storage or for people expecting cloud-scale model quality.
- Visible result
- A model responds locally, and you can identify where its files live and how much disk space it used.
- Time
- 45 min
- Device
- A supported computer; more memory allows larger models
- Account
- No third-party account expected for the basic local path
- Cost
- Software is open source; electricity, storage, and optional services still have costs
- Data
- Several gigabytes of free disk space; exact need depends on model
- HAACC check
- Editorially reviewed
Not run by HAACC
Before you start
A supported computer; more memory allows larger models
No third-party account expected for the basic local path
Software is open source; electricity, storage, and optional services still have costs
Several gigabytes of free disk space; exact need depends on model
Comfort following official install steps and checking storage
The path
- 01
Check your machine first
Record your operating system, available memory, and free disk space. Choose the smallest model recommended in current official documentation, not the most popular large model.
Continue only whenYou know the model download size before starting it. - 02
Install only from the official source
Follow the current operating-system instructions from the repository or official site. Decline unrelated installers and do not run commands from copied social posts.
Continue only whenThe local Ollama service starts without requesting unrelated permissions. - 03
Download one small model
Use the current official model library to select one small model. Note its name and size so you can remove the correct files later.
Continue only whenThe model appears in the local model list and disk usage increases by the expected amount. - 04
Run a bounded test
Ask three simple questions that contain no personal or confidential data. Save one response and note the response time and any obvious error.
Continue only whenYou receive a response while disconnected from any optional cloud integration.
One selected model answers locally and its storage footprint is known.
Evidence boundary
HAACC reviewed the public source and shaped this path, but did not run the project on a clean machine. Treat every install step as unverified.
What HAACC checked
- Repository identity, snapshot metrics, declared MIT license, and the local-model product concept
- A first-use path that begins with capacity and removal planning
What remains unchecked
- Installation and successful inference on any specific hardware
- Model quality, network behavior, current model licenses, and exact storage use
Stop and read
Models can exhaust storage and memory
Start small and watch resource use. The application license does not automatically cover every downloaded model.
Local does not mean automatically private
Optional integrations, logs, or other software can still move data. Test with non-sensitive text first.
When something fails
The model download stops or storage becomes low
Cancel, remove the incomplete or selected model using current official instructions, then choose a smaller model.
The computer becomes unresponsive or extremely slow
Stop the run. Do not keep retrying a model that exceeds your machine's practical capacity.
Stop the local service, remove downloaded models with the current official method, then uninstall the application and confirm disk space is reclaimed.
Turn it into your project
Offline note explainer
Turn one local text file into a repeatable summary workflow without sensitive material.
Small-model fit report
Publish response time, memory use, quality limits, and removal evidence for one machine.
Source snapshot
- Repository
- ollama/ollama
- Captured
- 2026-07-20
- Stars at capture
- 176,451
- Forks at capture
- 17,031
- GitHub license field
- MIT
- Pinned revision
- Not pinned