Ollama is a free app that runs open models on a Mac, a Windows PC or a Linux computer. Connect it to SteerClaw and the model on your computer powers the agent on your phone, with the conversation staying between the two devices.
Step 1: Install Ollama and download a model
- Install Ollama. Download it from ollama.com/download. There are installers for macOS, Windows and Linux.
- Download a model that supports tool calling. Open a terminal and run
ollama pull qwen3.8:27b. Tool calling lets the model ask SteerClaw to open a file or run a search. - Check that it answers. Run
ollama run qwen3.8:27b, type a question, and type/byeto leave. Ollama is now serving on port 11434.
We recommend these three models for agent work. All of them use tools and accept images.
| Model | Ollama name | Download | Why |
|---|---|---|---|
| Qwen 3.8 27B | qwen3.8:27b |
18 GB | The one to try first |
| Ornith 1.5 35B-A3B | ornith-1.5:35b |
23 GB | Fast for its size: only 3B of its 35B parameters work on each word |
| Qwen 3.8 Flash Next | qwen3.8-flash-next:125b-a6b-q4_K_M, or qwen3.8-flash-next:125b-mlx on Apple silicon |
120 GB / 105 GB | The most capable, for high-end machines |
Try others too. Any model marked tools at ollama.com/search?c=tools can use SteerClaw’s tools.
Step 2: Let Ollama accept connections from your phone
Out of the box, Ollama only accepts connections from the computer itself (127.0.0.1), so your phone cannot reach it.
macOS and Windows. Open Ollama’s Settings and turn on Expose Ollama to the network.
If your version of Ollama does not have that switch, set OLLAMA_HOST to 0.0.0.0:11434 instead:
-
On a Mac, run this in Terminal, then quit and reopen Ollama. It lasts until the Mac restarts.
launchctl setenv OLLAMA_HOST "0.0.0.0:11434" -
On Windows, quit Ollama from the taskbar. Open Settings (Control Panel on Windows 10), search for “environment variables” and choose “Edit environment variables for your account”. Add a variable named
OLLAMA_HOSTwith the value0.0.0.0:11434, click OK and start Ollama again.
Linux. If you installed Ollama with its install script, it runs as a background service. Run sudo systemctl edit ollama.service, add these two lines where the editor’s comments say to, then save:
[Service]
Environment="OLLAMA_HOST=0.0.0.0:11434"
Then run sudo systemctl daemon-reload && sudo systemctl restart ollama.
The first time, your computer’s firewall may ask whether Ollama can accept incoming connections. Allow it. On Windows, allow it on private networks, and check that your home Wi-Fi is set to private (Settings, Network & internet, Wi-Fi, your network, Network profile type). Windows sets new networks to public, which blocks the phone.
Step 3: Add Ollama in SteerClaw
- Open Settings and tap AI Models. Under Add new models, tap Your own server. My own server is already selected.
- Enter the server address. Type the computer's address and Ollama's port, for example
192.168.1.10:11434. SteerClaw addshttp://and/v1; the port is needed. To name the connection, open Server options. - Tap Connect and approve the local connection. SteerClaw warns that traffic to a local address is not encrypted. On your own home network, tap Allow & discover. On iPhone, also allow SteerClaw to find devices on your local network when iOS asks.
- Choose a model. SteerClaw lists the models you downloaded in Ollama and what each one supports. Pick
qwen3.8:27bor whichever you downloaded.
To test it, drop a PDF into the connected folder and ask “Summarize the PDF in my folder”. You will see the agent read the file before it answers.
Using Ollama away from home
The local address only works while the phone is on the same Wi-Fi as the computer. From mobile data or another network, use Tailscale: install it on both devices and enter the computer’s Tailscale address in SteerClaw instead. The Tailscale guide covers it step by step. Ollama needs no extra settings.
Tips
- Keep the computer awake. Ollama cannot answer while the computer sleeps. Adjust the energy settings, or on a Mac run
caffeinate -iin Terminal and leave the window open. Closing a laptop’s lid still puts it to sleep. - Give the model room to work. Long documents and web pages fill the model’s working memory, its context length. Ollama sets a default based on your graphics memory, often only 4K tokens, and recommends at least 64K for agents. Raise it with the Context length slider in Ollama’s Settings, or with
OLLAMA_CONTEXT_LENGTHon Linux. Do this before adding the model in SteerClaw, and run the model once first, so SteerClaw reads the size Ollama actually uses. - The first reply after a pause is slower. Ollama loads the model on the first request and keeps it in memory for 5 minutes.
Frequently asked questions
Which Ollama models do you recommend for SteerClaw?
For agent work we recommend Qwen 3.8 27B (qwen3.8:27b), Ornith 1.5 35B-A3B (ornith-1.5:35b) and, on high-end machines, Qwen 3.8 Flash Next (qwen3.8-flash-next). All three use tools and accept images. Try others too: any model marked tools at ollama.com/search?c=tools can use SteerClaw's tools.
Do I need an API key for Ollama?
No. Ollama does not use keys, so leave the access key field empty. If you put Ollama behind a proxy that asks for a key, enter it under Server options.
Does my data leave my computer?
The model runs on your computer, so prompts and the files you attach go from the phone to the computer and no further. There are three exceptions. Web searches go to the search provider you chose in SteerClaw, and pages the agent reads are fetched from their sites. Work tools you connect, such as GitHub or Notion, receive what the agent sends them. Ollama models whose names end in -cloud run on Ollama's servers.
Can I use Ollama away from home?
Yes, with Tailscale, which gives the computer an address your phone can reach from any network. The Tailscale guide walks through it. Without it, the local address only works on the same Wi-Fi.