Getting started

TrailSail is a native macOS chat client for the models running on another Mac of yours. This page takes you from a download to your first streamed reply.

The two parts

  • TrailSail, the app, is the harness. It runs the loop around the model: it sends the chat, carries out the tool calls, asks your permission, keeps the context in budget and stores your chats. Tools run on the Mac the app is on.
  • trailSailServer, the package, is the engine manager. It installs the engines on the other Mac, keeps them running, switches and downloads models, and makes the keys. Chats go straight from the app to the engine and never pass through it.

What you need

  • Two Macs, or one: see Everything on one Mac. The one you work on: macOS 14 or later, on Apple Silicon. The one that runs the models: Apple Silicon, at least 16 GB of memory (at least 32 GB for 27B models) and at least 40 GB of free disk. These are minimums. Splash, the faster of the two engines, needs an Apple M3 or newer and macOS 26.4 or later; on an older Mac the package installs the llama.cpp router alone.
  • A Tailscale account, with Tailscale installed and signed in on both Macs. See the next section.
  • On the server: Homebrew, and OrbStack or Docker Desktop if you want web search.

Set up Tailscale first

Tailscale is the private network that joins your two Macs, wherever they are. TrailSail itself has no account and no sign-up, but Tailscale does, and the server package will not install without it.

  1. Make a Tailscale account at tailscale.com. It is free for personal use. For use at a company, check Tailscale's pricing.
  2. Install Tailscale on both Macs from tailscale.com/download and sign in to the same account on each. Both Macs then show in the Tailscale menu as connected.
  3. Turn on HTTPS, once for your account. In the Tailscale admin console, open DNS, check that MagicDNS is on, and press Enable HTTPS under HTTPS Certificates. The server package publishes the engines over HTTPS inside your Tailscale network, and Tailscale needs this switch to do that.

Nothing is opened to the internet: the server is reachable only by devices signed in to your Tailscale network.

Three steps

  1. On the server, download trailSailServer-<version>.pkg and open it. Enter your password once. A dialog shows a six-word pairing code; it is also on the clipboard.
  2. On your Mac, download TrailSail, open the disk image and drag the app to Applications. Open it, then go to Settings (⌘,) → Server → Pair with a server. Enter the server's Tailscale name (for example mini.tail1234.ts.net) and the code.
  3. Start a chat. The sidebar's profile menu lists the server's engines; the first reply loads the model.

Everything on one Mac

You do not need a second Mac. The models can run on the Mac you work on, as long as it has the memory for both: the model takes its share, and macOS, TrailSail and your other apps need the rest. A 16 GB Mac runs small models this way; 32 GB or more is comfortable.

There are two ways to set it up.

Without the server package

The simplest: run a model server yourself and point TrailSail at it. Nothing leaves the Mac, so Tailscale is not needed. Two servers to choose from:

Splash is the fastest on Apple silicon, and the engine the server package uses. It needs an Apple M3 or newer and macOS 26.4 or later.

  1. Install it and start it with a model. The first start downloads the model; wait for the line that says Ready, and leave that Terminal window open.

    brew install incoai/tap/splash
    splash serve --model mlx-community/Qwen3.8-27B-4bit --max-memory 20G
    

    --max-memory caps what Splash may use, so the rest stays with your other apps: 20G suits a 32 GB Mac. Splash's own guide asks for 36 GB or more for this model and lists smaller models for Macs with less.

  2. In TrailSail, open Settings → Connections → Add and fill in:

    • Name: anything, for example This Mac
    • Base URL: http://localhost:8000
    • API key: leave empty. Splash on its own Mac needs none.
    • Server type: Splash
  3. Press Test connection. The model it lists confirms it works. Save, and start a chat.

Ollama runs on any Apple Silicon Mac.

  1. Install Ollama and download a model with it that fits your memory.
  2. Add it the same way, with Base URL http://localhost:11434, no API key, and Server type Ollama.
  3. Press Test connection, save, and start a chat.

Any other OpenAI-compatible server on the same Mac works the same way, with its own address. You look after the model server yourself; the Server panel is for the server package only.

With the server package

The package can also be installed on the Mac you work on. You get the Server panel, the installer and the engines it manages, exactly as on a separate server.

  1. Set up Tailscale on this Mac, with HTTPS turned on, as in Set up Tailscale first. The package publishes its engines through Tailscale even when the app is on the same Mac.
  2. Open trailSailServer-<version>.pkg on this Mac. It ends with a pairing code.
  3. In TrailSail, go to Settings → Server → Pair with a server and enter this Mac's own Tailscale name and the code.
  4. Open Settings → Server and look at Memory for models. TrailSail notices the server is the Mac you are on, recommends a lower figure than it would for a separate server, and warns if the figure is too high. See Memory for models.

One thing to know: the package starts its main engine when you log in and keeps a model loaded. When you need the memory for something else, stop the engine from the Server panel or from the sailboat in the menu bar, and start it again when you want to chat.

With a server you already run

Any OpenAI-compatible server on another machine works too: Settings → Connections → Add, paste the URL and key, then Test connection. Tailscale is only needed if you reach that machine through it. See Manual setup for Splash, llama.cpp, Ollama and Strata by hand.

Then, in rough order

You wantGo to
Everything the package installs and how it runsServer setup
The six-word code, in detailPairing
Run, switch and manage the machine from the appThe Server panel
The same, as commands on the serverThe trailSailServer CLI
A folder the model can read, search and editProjects
Permission tiers for the developer toolsDeveloper tools
Parallel sub-agentsAgents
Markdown memory and wrap-upNotes & memory
Your setup to move between MacsSync

Found a problem, or is something here unclear? Report it.