CodeBridge: coding-agent answers taught out loud, powered by Vocal Bridge

One command turns the last thing Claude Code or Codex said into a live voice lesson with a board that visualizes each idea as the tutor teaches it.

The problem: coding agents accelerate output, not understanding

Every developer using a coding agent hits the same moment. The answer is good enough to apply, so you skim it, accept the patch, and keep moving. The code advances, but your mental model does not. Repeat that several times a day and the agent creates software faster than it creates comprehension.

The usual fallback is still more text. Ask the agent to simplify the answer, scan another explanation, or open the linked file and try to connect it yourself. That adds reading without fixing the moment that matters, the point where confusion appears and you need to ask why.

What Taughtful.ai built with CodeBridge

CodeBridge closes that gap with a command inside the developer's existing workflow. It takes the last answer from Claude Code or Codex and turns it into the first section of a live lesson at Taughtful.ai. Any files explicitly tagged with @ become additional sections, including code, PDFs, Office documents, ebooks, screenshots, and photos of handwritten notes.

This is not text-to-speech over the original response. The tutor teaches the material again in its own words, one spoken beat at a time, while the board picks a visual shape that fits the idea. That might be a flow diagram, comparison, misconception card, analogy map, timeline, chart, code block, or a set of parallel cards. The user can interrupt at any point, ask for an analogy grounded in cricket or finance, or switch languages while keeping technical terms in English. At the end, CodeBridge can write a ready-to-run prompt that carries the lesson back into the coding session.

Taughtful.ai's supplied launch documentation says the CLI sends the selected answer and explicitly tagged files, screens out files that look like secrets (with .env, keys, and credentials given as examples), and declines answers under 200 words when no files are attached. The same documentation describes the CLI as MIT-licensed and telemetry-free, and says the command uses no coding-agent model tokens. These are Taughtful.ai's product claims, not an independent security or privacy audit by Vocal Bridge.

How CodeBridge uses Vocal Bridge

CodeBridge uses Vocal Bridge's Voice as a tool call surface. The lesson begins from another agent interface, then opens a purpose-built voice experience at Taughtful.ai. Vocal Bridge powers the real-time conversation once the lesson starts, including low-latency voice, natural turn-taking, and barge-in when the learner interrupts mid-explanation.

"The hardest part of a voice tutor isn't the talking, it's the stopping. Vocal Bridge gave us a tutor you can cut off mid-sentence, and with client actions the same agent renders slides, focuses sections, and hands you a prompt while it speaks. Voice in, control out, one loop. That would have been months for us to build; on Vocal Bridge it was days."
- Sapna Sangmitra, Founder of Taughtful.ai

The spoken lesson and the visual board work as one interface. As the tutor moves to a new idea, Taughtful.ai can render the matching diagram or comparison on screen through client actions, so the learner hears and sees the same beat rather than watching a static transcript.

import { useEffect, useState } from "react";
import { useAgentActions } from "@vocalbridgeai/react";

export function LessonBoard() {
  const [currentVisual, setCurrentVisual] = useState(null);
  const { onAction } = useAgentActions();

  useEffect(() => {
    const unsubscribe = onAction("show_lesson_visual", setCurrentVisual);
    return typeof unsubscribe === "function" ? unsubscribe : undefined;
  }, [onAction]);

  return <pre>{JSON.stringify(currentVisual, null, 2)}</pre>;
}

Illustrative. Taughtful.ai supports a richer visual vocabulary and section-aware lesson flow.

The architectural split is clean. The CodeBridge CLI owns local session capture, explicit file selection, the secret-file screening described by Taughtful.ai, and the handoff into Taughtful.ai. Taughtful.ai owns lesson planning, visual selection, document and image ingestion, learner preferences, and the prompt produced at the end. Vocal Bridge owns the live voice layer and the client-action channel that keeps the tutor and the board in sync.

What's hard about voice here

• Interruption is the main interaction. A learner asks “why?” at the exact point an explanation stops making sense, often while the tutor is still speaking. The tutor has to stop, preserve the context, and rebuild the lesson around the question. Vocal Bridge's barge-in and turn-taking handle that interruption.
• The board cannot drift from the voice. A diagram that updates too early or lingers on the previous concept makes the lesson harder to follow. Vocal Bridge client actions let Taughtful.ai pair each spoken beat with the visual that belongs to it.
• Lesson work cannot become dead air. Rebuilding an explanation, switching languages, or moving through an attached file can take a beat. Vocal Bridge's real-time streaming keeps the tutor present while Taughtful.ai prepares what comes next.