Over the past few weeks, I shipped a new macOS tool called Lectio.
The starting point is simple: when you are reading a document, course, paper, product guide, or error message, the moment of confusion is usually small and immediate. Maybe one sentence does not make sense. Maybe a term is unclear. Maybe a stack trace needs one more explanation. Maybe a paragraph depends on context you have not fully connected yet.
In theory, this is exactly where AI should help. In practice, the workflow is often clumsy. You switch to a chat app, copy text, paste context, explain what you were reading, then ask the actual question. By the time you finish that setup, the original question has already lost some of its shape.
Lectio is meant to remove that friction.
Not Another General Chat Window
I do not want Lectio to become yet another general-purpose chat tool. There are already many of those, and many of them are excellent. Lectio is intentionally narrower. It focuses on one situation:
I am looking at something, I am stuck, and I want to ask about what is already in front of me.
That makes the interaction deliberately simple:
- You are reading something.
- If there is a specific passage, select it.
- If nothing is selected, Lectio reads the visible screen content.
- You ask your question out loud.
- The answer stays in a floating panel, and you can listen to it as speech.
This is a small workflow, but it matters. For reading, the key is not whether the AI can produce an impressive answer in isolation. The key is whether it understands the exact piece of material you are looking at right now.
Why Voice
When we genuinely do not understand something, the most natural response is often not typing. It is saying:
“What does this sentence mean?”
“Why is this designed this way?”
“Can you give me another example?”
Voice is closer to the first draft of thought. Typing forces the question to become more polished. That can be useful in some contexts, but in the moment of reading friction, it often becomes one more obstacle.
Lectio is not meant to replace deep note-taking or full research tools. It is closer to a study companion sitting next to you. You can ask an imperfect question first, then follow the answer into the next question.
The Current Screen Is the Context
Lectio’s context strategy is straightforward:
- Selected text comes first. This is useful for a precise paragraph, function, error message, or sentence in a document.
- When nothing is selected, Lectio reads the current screen. This works across browsers, PDFs, course players, settings pages, and error screens.
- Answers stay grounded in the material at hand. It does not start by assuming you want to search the whole web. It starts by understanding what you are already reading.
I think this is an important direction for AI learning tools. Many learning moments do not fail because answers are unavailable. They fail because the tool does not know the context sitting right in front of the learner.
Trust Boundaries
Lectio also has a deliberate trust boundary.
Screen reading, speech input, and spoken output are handled locally through macOS capabilities where possible. What gets sent to the AI service is text content and your question, not an unnecessary full-screen screenshot.
The AI service key is also provided by the user. That means you keep control over the provider, account, cost, and data path. It adds one setup step, but I think it is the right direction for this kind of early product.
Where It Is Now
Lectio is currently an early beta for Apple Silicon macOS. The latest shipped version is v0.0.4.
This version includes:
- selected-text-first context reading
- current-screen reading when no text is selected
- voice questions and spoken responses
- a floating conversation panel
- bring-your-own AI service key setup
- local macOS capabilities for screen and speech-related workflows
You can view the project page here:
You can also download the early beta directly:
What Comes Next
Lectio is still early, but I already like the direction. It is not a large system. It is a small tool close to everyday learning. The problem it solves is not “can AI answer questions?” The problem is “when I get stuck, can there be less switching, less setup, and less friction?”
Next, I want to keep improving:
- more stable screen understanding
- more natural follow-up context
- smoother first-run setup and key management
- more real-world testing across technical docs, online courses, and PDFs
If OmniTypist helps me get thoughts into the computer faster, Lectio helps me get the material in front of me into my head faster.
One is for output. One is for intake.
Together, they are starting to form the workflow I want: using AI to reduce the friction of knowledge work without outsourcing the thinking itself.
Comments & Feedback