Open source matters for AI tools in a practical way: you can inspect what they send, run them offline, and fork them when your needs change. These six cover local models, dictation, cloud management, agent interfaces and LLM infrastructure.
For running models locally
Ollama is the most widely used way to run open-weight language models on your own machine: one command downloads and runs Llama, Qwen, DeepSeek and community variants. Prompts are processed locally rather than sent to a provider, and the software itself is free - the real cost is your hardware and memory.
For private dictation on Mac
BetterWispr is free, open-source dictation for macOS where everything runs on-device: Apple, Parakeet and Whisper models download once and work offline, with no account and no cloud upload. It types polished text into any app, matched to the writing style you set per app.
For a dictation tool you can fork
Handy is a minimal open-source push-to-talk for any text field, running fully offline on Whisper across platforms. Its codebase is deliberately small and hackable, built for tinkerers and accessibility use cases that commercial tools ignore.
For cloud management without the console
Lazy AWS is an open-source, keyboard-driven terminal UI for browsing and acting on AWS resources, using your existing AWS CLI v2 credentials. Runs on macOS and Linux; a fast answer for developers who live in the terminal.
For agents that set up their own tools
Craft Agents is an open-source interface for Claude, ChatGPT, OpenRouter and local models whose agents configure their own integrations: ask it to add Linear or Slack and it finds the API or MCP server, reads the docs and wires itself up. Connect via browser, API, MCP or local filesystem.
For production LLM infrastructure
TensorZero is an open-source stack for industrial-grade LLM apps: a unified gateway to every major model provider with sub-millisecond latency, plus observability for inferences, metrics and human feedback. Built for teams standardizing on infrastructure they can inspect and self-host.
Allow Google Analytics to help us understand how LaunchPerk is used? Optional analytics stays off unless you accept.