Technical Guide to GitNeural and Visual AI Tutoring Workflows

Practical Overview & Architecture

Knowable is marketed as the world's first visually grounded AI tutor, designed specifically to provide real-time feedback using on-paper problem-solving progress. Rather than functioning as a standard text-based chatbot or automated answer-dispensing application, the system relies on an integrated architecture that connects a local Mac client with frontier vision models hosted on cloud infrastructure. When a student interacts with the application, camera frames capture the physical textbook or notebook page in real time. These inputs stream over secure protocols directly to AWS Bedrock, specifically residing in the us-east-1 region, where the foundational models perform visual inference.

The underlying software architecture deliberately avoids storing or sharing raw camera footage. Camera frames exist exclusively in memory during an active study session and are never written to permanent disk storage or database instances. Persistent data is strictly limited to chat messages and session metadata, such as timestamps, event logs, and hint counts, which are stored securely using Amazon DynamoDB with encryption applied both in transit and at rest. This design choice allows multi-mac session continuity while strictly protecting user privacy. Furthermore, the infrastructure relies on contractual guarantees from cloud providers ensuring that user inputs and model outputs are never used to train the underlying foundation models or exposed to external entities.

Prerequisites & Installation/Setup

To successfully deploy and utilize the Knowable application, specific hardware and operating system prerequisites must be met. The software requires a compatible Mac running macOS Ventura or a later version. No specialized external hardware or complex configuration steps are necessary, as the application leverages the built-in Mac webcam operating in a specialized Desk View mode. Alternatively, users can utilize an iPhone via Apple Continuity Camera functionality to capture their desktop workspace. The notebook or textbook must be positioned physically on the desk in front of the camera feed so that the vision models can accurately interpret diagrams, equations, handwriting, and printed text across any high-school subject.

Getting the application up and running requires minimal setup time, generally advertised as taking approximately three minutes. Users begin by aiming their integrated webcam or connected iPhone camera directly at their study materials. Interaction with the AI tutor is initiated through straightforward desktop mechanics: pressing and holding the opt+M keyboard combination activates push-to-talk voice input, allowing the student to speak their question naturally while the AI simultaneously analyzes the live camera frame captured at that exact moment. Alternatively, users can type their queries directly into the chat interface. The system then processes the input, renders mathematical notation using SwiftMath for LaTeX output, and delivers synthesized speech via text-to-speech engines.

Documented Implementation Workflow

The documented workflow for students centers around active problem-solving without receiving direct answers. When a student encounters a difficult concept in subjects such as mathematics, science, English, history, or foreign languages, they position their notebook under the camera view. By holding the opt+M shortcut, the student submits a voice query or types a question into the text box. The underlying vision models evaluate the exact camera frame alongside the user query, generating a pedagogical response. Rather than solving the equation or writing out the final solution, Milo provides a guiding question designed to help the student deduce the answer independently. This mechanism ensures that the conceptual understanding is properly reinforced before examinations.

For educators and administrators, a parallel implementation workflow is accessible via the official educator portal located at platform.knowable.ca. Teachers can create a dedicated class environment, generate an invite code, and distribute it to students so that Milo aligns its coaching directly with the specific curriculum being taught in class. The educator portal surfaces valuable analytics highlighting which specific concepts students struggled with prior to classroom sessions. Additionally, instructors and administrators maintain tri-state student sharing controls, allowing them to decide precisely how much insight is shared back with parents, ranging from disabled visibility, stats-only reporting, or comprehensive stats combined with complete session activity tracking.

Known Limitations, Tradeoffs & Error Scenarios

Despite its advanced capabilities, Knowable introduces specific technical limitations and tradeoffs that users must account for during deployment. The application is strictly restricted to macOS environments, meaning users on Windows, Linux, or mobile operating systems outside of iOS Continuity Camera integration cannot run the desktop client locally. Additionally, the real-time processing of video frames, audio transcription via push-to-talk, and complex LaTeX-rendered math replies require consistent network connectivity to AWS cloud endpoints. If network latency spikes or connectivity drops, streaming interruptions can impact the responsiveness of the text-to-speech and vision inference pipelines.

Another structural limitation involves the deliberate design choice regarding hint generation. Because the system is purposefully engineered to withhold direct solutions and instead offer guiding questions, students seeking quick answer verification or copy-paste homework completions will experience friction. Furthermore, usage is governed by credit tier allocations. Free accounts receive a recurring allocation of 1,000 credits each month, while Knowable Plus subscriptions grant 10,000 credits per month. Heavy users who exhaust these allocations must rely on one-time in-app credit purchases, such as 500 credits for $4.99 USD or 3,000 credits for $24.99 USD, which do not expire but require active App Store account management for billing and subscription cancellations.

Who Should Use It & Production Fit

Knowable is best suited for high school students, parents, and educators looking for an interactive, visually grounded study aid that fosters genuine comprehension rather than rote memorization. High school students across any academic discipline—ranging from complex calculus and physics diagrams to textual analysis in literature—benefit from having an AI tutor available 24/7, particularly during late-night study sessions when human tutors are unavailable. Parents looking for affordable alternatives to expensive human tutoring sessions find value in the platform's structured pricing tiers and transparent data privacy practices, which guarantee that camera feeds are never stored or shared with advertisers.

Educators managing high school curricula also represent an ideal user base, as the platform integrates smoothly into classroom instruction via the educator portal. By leveraging curriculum-aligned coaching and reviewing pre-class struggle analytics, teachers can optimize their lesson plans. Individuals seeking a zero-risk evaluation can begin with the free tier, which provides 1,000 credits monthly without requiring a credit card, while power users and families can opt into the Knowable Plus subscription via the Mac App Store for expanded monthly credit limits and continuous access across academic subjects.