AI Productivity Suite: Desktop Workflow Automation

01 Challenge

In day-to-day operations (support, content), context switching costs time: copy text, open ChatGPT, paste, wait, copy back – and sensitive data ends up in the cloud.

02 Approach

A native GTK3 tool launches with the clipboard content, offers predefined actions and streams the response of a local Ollama model in real time.

03 Result

AI functions at a keystroke, right inside the desktop workflow – no copy-paste detours, and no company data leaving the laptop.

Stack: Python 3GTK3 / PyGObjectOllama REST APIrequeststhreading

In the day-to-day operations of an e-commerce manager (e.g. customer support, content creation), a lot of time is lost through context switching: you copy text, open ChatGPT, paste it, wait, copy it back.

Ollama Menu is a “productivity suite” that integrates AI functions directly into the desktop workflow. It eliminates the “copy-paste madness” and gives employees access to company-owned AI models with a single keystroke.

Default actions: 6; Summarize · Pro/Con · Translate · Explain · Correct
History entries: 100; Input, output, model, action
Inference: local; any Ollama model selectable
Instance: 1; Single instance via PID file

Ollama Menu screenshot: two windows with clipboard input, action bar, model selection and AI response — Fig. 1 — Ollama Menu in action: “Summarize” with expanded history on the left, “→ English” translation on the right – both with gemma3:4b

Business Value: Efficiency Gains

Before: copy-paste workflow

Copy text
Open ChatGPT in the browser
Paste, type a prompt, wait
Copy the answer back
Data leaves the company

With Ollama Menu

Tool launches with the clipboard
Pick an action with the arrow keys
Ctrl+Enter – the answer streams live
Use the result right away
Local model, data stays on the laptop

Support acceleration: An agent marks a customer email and selects “Draft Reply” – the draft is instantly in the clipboard.
Content standardization: Writers use predefined prompts (e.g. “SEO optimization for product description”) for consistent results.
Privacy: Since local models (via Ollama) are used, no sensitive company data ever leaves the employee’s laptop.

Key Features

Seamless clipboard integration: Launches with the current clipboard content, which can be edited directly.
Customizable actions: An external prompts.json defines actions with their own system prompts – freely extensible.
Dynamic model selection: Automatically detects all locally available Ollama models and lists them in a dropdown.
Live streaming: The response appears in real time as it is generated – key for the user experience with longer answers.
History: Saves the last 100 requests (input, output, model, action) for quick restoring and copying.
Keyboard control: Arrow keys to choose an action, Ctrl+Enter to submit, Esc to close.

Architecture

The tool is built as a native GTK3 interface to consume minimal resources and run stably in the background.

01InputClipboard content, editableClipboard
02Action & modelSystem prompt from prompts.json, model from Ollamaprompts.json
03Background threadRequest to the Ollama REST API, previous request is cancelledthreading
04StreamingUI updates safely on the main threadGLib.idle_add
05HistoryResult is saved and can be copiedhistory.json

Fig. 2 — Lifecycle of a request

Technical Highlights and Solutions

Responsive UI through Multithreading

Problem: Requests to the Ollama API can take several seconds. Run on the main thread, they would freeze the entire interface.

Solution: Each request runs in its own background thread (threading.Thread), so the GUI stays responsive at all times. To safely update components like the response text field from the background, GLib.idle_add() is used: all UI updates run on the main thread, preventing race conditions.

Robust Process Management

Problem: A user might accidentally start a new request while an old one is still running.

Solution: Cancellation logic using threading.Event: before a new thread starts, the system checks whether one is already running – if so, it receives a cancel signal. This ensures predictable behavior and prevents overlapping requests. Live streaming also gives immediate feedback.

Clean Architecture and Configuration

Problem: Hardcoded prompts and settings make an application inflexible.

Solution: The application follows the XDG Base Directory standard: configuration (prompts.json) lives in ~/.config, application data (history.json) in ~/.local/share. Code and user data are separated, keeping the tool easy to customize and maintain.

Professional Application Handling

The single-instance lock via a PID file (/tmp/ollama-menu-python.pid) prevents users from accidentally opening multiple windows, which would cause confusion and unnecessary resource consumption.

Structured logging (the logging module) and clean exception handling ensure stability and make debugging easier.