Skip to main content

Ollama Manager

A terminal UI (TUI) for managing Ollama models.

Why a TUI?​

The command line is powerful but repetitive. Instead of typing:

ollama ps          # What's loaded?
ollama list        # What's available?
ollama run qwen3   # Load this one
ollama stop qwen3  # Unload it

Just run ollama-manager.exe and use keyboard shortcuts.

Installation​

Pre-built Binary​

Download from the repository:

cd ollama-manager
.\ollama-manager.exe

Build from Source​

Requires Go 1.21+ or Docker/Podman:

# Using container (no local Go needed)
cd ollama-manager
.\build.ps1

# Using local Go
.\build.ps1 -Local

Usage​

Starting the Manager​

.\ollama-manager.exe

You'll see a list of all installed models with their status:

┌──────────────────────────────────────────────────────────┐
│  Ollama Manager                                          │
├──────────────────────────────────────────────────────────┤
│  > qwen3:32b                              [LOADED]       │
│    deepseek-r1:32b                                       │
│    llama3.3:70b-instruct-q4_K_M                          │
│    llama3.1:8b                                           │
│    mistral:7b                                            │
│    phi-4:14b                                             │
│                                                          │
├──────────────────────────────────────────────────────────┤
│  r: run  s: stop  u: unload all  R: refresh  q: quit    │
└──────────────────────────────────────────────────────────┘

Keyboard Controls​

KeyAction
↑ / ↓Navigate models
r / EnterRun selected model (interactive chat)
sStop selected model (unload from VRAM)
uUnload ALL models
RRefresh model list
qQuit

Running a Model​

  1. Navigate to a model with arrow keys
  2. Press r or Enter
  3. The TUI exits and opens an interactive chat session
  4. Type /bye to exit chat

Stopping Models​

Models stay loaded in VRAM for fast reuse. To free memory:

  1. Navigate to a loaded model (shows [LOADED])
  2. Press s to stop it
  3. Or press u to unload ALL models

How It Works​

The manager is built with:

It calls Ollama CLI commands:

// List models
exec.Command("ollama", "list")

// Check what's loaded
exec.Command("ollama", "ps")

// Run model
exec.Command("ollama", "run", modelName)

// Stop model
exec.Command("ollama", "stop", modelName)

Source Code​

The full source is in ollama-manager/main.go:

package main

import (
    tea "github.com/charmbracelet/bubbletea"
    "github.com/charmbracelet/lipgloss"
)

type model struct {
    models   []modelInfo
    cursor   int
    selected string
}

// ... see full source in repository

Building​

No local Go installation required:

cd ollama-manager
.\build.ps1

This:

  1. Builds a Docker image with Go
  2. Cross-compiles for Windows
  3. Extracts the .exe file

With Local Go​

If you have Go installed:

cd ollama-manager
.\build.ps1 -Local

Or manually:

go mod tidy
go build -ldflags="-s -w" -o ollama-manager.exe .

Build Output​

Build complete! Binary: ollama-manager.exe (2.57 MB)

Customization​

Adding Features​

Fork the repository and modify main.go:

// Example: Add model deletion
case "d":
    exec.Command("ollama", "rm", m.models[m.cursor].name).Run()

Changing Styles​

Modify the Lipgloss styles:

var (
    titleStyle = lipgloss.NewStyle().
        Bold(true).
        Foreground(lipgloss.Color("86"))  // Change color
)

Troubleshooting​

"ollama not found"​

The manager calls ollama CLI. Ensure it's in your PATH:

ollama --version

Models not showing​

Refresh the list with R key.

"Access denied" on Windows​

Run PowerShell as Administrator if models are in a protected directory.

Terminal rendering issues​

Ensure your terminal supports ANSI colors. Windows Terminal recommended.

Alternatives​

If a TUI isn't your style:

CLI Commands​

ollama list    # List models
ollama ps      # Show loaded
ollama stop X  # Unload model

Open WebUI​

Web interface at http://localhost:3000 with visual model selection.

API​

# List via API
curl http://localhost:11434/api/tags

# Load model via API
curl http://localhost:11434/api/generate -d '{"model":"qwen3:32b"}'