Skip to content
Tallyfy

Connect Ollama to Tallyfy Desktop

What gets installed

Section titled “What gets installed”

Tallyfy Desktop drives Ollama through the local daemon at http://127.0.0.1:11434. Unlike the other three providers, Ollama isn’t a CLI you spawn per-run; it’s a background service that listens on a port and accepts HTTP requests. Tallyfy Desktop talks to that service directly.

The install command per platform is baked into the desktop app source code:

PlatformCommand
macOSbrew install ollama && brew services start ollama
Linuxcurl -fsSL https://ollama.com/install.sh | sh
Windowswinget install --silent --accept-source-agreements --accept-package-agreements Ollama.Ollama

After install, Ollama runs as a background service that starts on login (macOS, Linux) or on boot (Windows). It listens only on 127.0.0.1, so other machines on your network can’t reach it.

Prerequisites

Section titled “Prerequisites”
  • A Tallyfy account
  • Enough RAM and disk space for the model you want to run (the default llama3.2:1b needs around 2 GB RAM and 1.3 GB disk)
  • Homebrew (macOS), curl (Linux), or winget (Windows)
Section titled “Option A: Install Wizard (recommended)”

  1. Open the Ollama tab

    Click the Ollama tab in Tallyfy Desktop. If Ollama isn’t installed or the daemon isn’t running, the Install Wizard appears.

  2. Read the consent screen

    The wizard shows the exact command per platform and lists the risks:

    • macOS: requires Homebrew; starts a launchd service homebrew.mxcl.ollama that listens on 127.0.0.1:11434 and auto-starts on login
    • Linux: creates a systemd unit ollama.service (requires sudo for service registration)
    • Windows: installs via winget (Windows Package Manager); the service listens on 127.0.0.1:11434

  3. Hold Shift for three seconds

    The button activates while you hold Shift.

  4. Watch the progress log

    The wizard streams installer output. On macOS you’ll see Homebrew’s download and link steps, plus the brew services start ollama line. On Linux you’ll see the install script’s output. On Windows you’ll see winget’s progress.

  5. Pull the default model

    After install, the wizard prompts you to pull a default model. The Tallyfy Desktop default is llama3.2:1b (1.2 billion parameters, around 1.3 GB on disk). The pull runs in the wizard with a progress bar.

    You can pull a bigger or smaller model later; this is just the cold-start choice.

  6. Test the connection

    Type hi in the Ollama tab and press Enter. You should see the streamed response within a few seconds (latency depends on your CPU/GPU).

Option B: Manual install

Section titled “Option B: Manual install”

  1. Install Ollama

    • macOS: brew install ollama && brew services start ollama
    • Linux: curl -fsSL https://ollama.com/install.sh | sh
    • Windows: winget install Ollama.Ollama

  2. Pull a model

    Terminal window
    ollama pull llama3.2:1b

    Or pick any other model from ollama.com/library ↗[1]. Larger models (llama3.1:70b, mixtral:8x7b, etc.) need substantially more RAM.

  3. Verify the daemon is running

    Terminal window
    curl http://127.0.0.1:11434/api/tags

    You should see a JSON response listing your installed models.

  4. Restart Tallyfy Desktop

    Quit fully and reopen. The Ollama tab will detect the daemon and the model.

First chat

Section titled “First chat”

Once Ollama is connected:

  1. Click the Ollama tab.
  2. The model picker shows the models you’ve pulled. Pick one.
  3. Type your message and press Enter.

First responses are slower because Ollama loads the model into RAM on the first request. After that, it stays loaded for a few minutes and responses are fast.

What Ollama can and can’t do in v1

Section titled “What Ollama can and can’t do in v1”

Ollama runs local open-source models, which have real limits compared to Claude, Codex, or Gemini:

CapabilityStatus
Streaming text responsesYes (NDJSON streaming)
MCP tool useDegraded (no native MCP; falls back to system-prompt prefix with Tallyfy context)
Cost reportingN/A (always 0; local compute)
Thinking eventsNo
File attachmentsNo
Browser automationNo
Tallyfy context injectionYes (via system-prompt prefix)
Task Intent WidgetYes

The Task Intent Widget still works because it watches the model’s text output for intent patterns; it doesn’t need MCP. So a small local model can still trigger a task creation when you ask for one.

Choosing a model

Section titled “Choosing a model”

Tallyfy Desktop defaults to llama3.2:1b. That’s a deliberate trade-off: small enough to run on any modern laptop, fast enough to feel responsive, but obviously less capable than larger models.

A few common choices, in increasing size and capability:

  • llama3.2:1b (~1.3 GB) - the default; very fast, modest reasoning
  • llama3.2:3b (~2 GB) - better at instructions, still fast on most hardware
  • mistral:7b (~4 GB) - solid general-purpose model, needs 8+ GB RAM
  • llama3.1:8b (~5 GB) - higher quality, slower; 16+ GB RAM recommended
  • qwen2.5-coder:7b (~5 GB) - tuned for code; useful if you’re using Ollama for development tasks

Pull any of these with ollama pull <model>, then pick it from the model picker. Larger models give better answers; smaller models give faster ones. There’s no subscription cost, just disk and RAM.

Privacy guarantee

Section titled “Privacy guarantee”

The whole point of Ollama is privacy. Tallyfy Desktop preserves this:

  1. Your prompts go from the desktop app to 127.0.0.1:11434 on the same machine. They never touch the internet.
  2. Model responses come back from the same local daemon.
  3. Tallyfy task metadata (provider=“ollama”, model=”…”, prompt summary, duration) gets sent to Tallyfy as a row in the “AI Runs” checklist. The same metadata as any other run.

If you want to run truly offline (no network at all, even for Tallyfy), the desktop app will queue the Tallyfy task creation and retry when network comes back. The AI run itself completes immediately.

Verifying the install yourself

Section titled “Verifying the install yourself”
Terminal window
# Is the binary on PATH?
which ollama


# Does it run?
ollama --version


# Is the daemon up?
curl -sf http://127.0.0.1:11434/api/tags || echo "daemon not running"


# What models are pulled?
ollama list

If the daemon isn’t running:

  • macOS: brew services start ollama (or brew services restart ollama if it’s wedged)
  • Linux: sudo systemctl start ollama
  • Windows: open Services, find “Ollama”, and start it

Then press Cmd+Shift+R (Ctrl+Shift+R) in the Ollama tab to retry detection.

Troubleshooting

Section titled “Troubleshooting”

”Daemon not running” but the binary is installed

Section titled “”Daemon not running” but the binary is installed”

The Ollama service might be stopped. Run the platform-specific start command above. If it still won’t start, check the logs:

  • macOS: ~/Library/Logs/Homebrew/ollama/
  • Linux: journalctl -u ollama -n 100
  • Windows: Event Viewer, Windows Logs, Application

Out-of-memory error mid-chat

Section titled “Out-of-memory error mid-chat”

You pulled a model bigger than your RAM. Switch to a smaller model from the picker, or pull a quantized variant (e.g., llama3.1:8b-q4_0 instead of llama3.1:8b).

Very slow first response

Section titled “Very slow first response”

Ollama loads the model from disk on the first request. For larger models this can take 10-30 seconds. Subsequent requests are fast. If you have a GPU (Apple Silicon, NVIDIA CUDA), Ollama uses it automatically; check ollama ps to see which device the model is loaded on.

”Model not found” when picking a model

Section titled “”Model not found” when picking a model”

You haven’t pulled it yet. From a terminal: ollama pull <model>. The download progress appears in the terminal; the model becomes available in Tallyfy Desktop once it finishes (you may need Cmd+Shift+R to refresh the picker).

Section titled “Related articles”
Get in touch
About Tallyfy