# An AI-operated audio workshop Legion project library, edition 2026-09-06-draft-01 (en). This project's own material: 5 authored documents, reproduced unchanged. Project page: https://legion.tw/projects/ai-apps/audio-workshop?edition=2026-09-06-draft-01 This document: https://legion.tw/api/projects/packet.md/ai-apps/audio-workshop.md?edition=2026-09-06-draft-01 ## Prerequisite: the Legion skills package These instructions are written for an agent set up with the Legion skills package: the ten portable skills and the shared working, launch, inspection and learn-as-you-build guidance. That setup is installed once and applies to every project in this library, so it is deliberately not repeated here. Skills package, complete Markdown: https://legion.tw/api/projects/skills-packet.md?edition=2026-09-06-draft-01 Skills package, human page: https://legion.tw/projects/skills?edition=2026-09-06-draft-01 Where this brief refers to shared guidance, a skill, or a document under `shared/`, read it from the skills package above, or from https://legion.tw/api/projects/doc/?edition=2026-09-06-draft-01. If the package is not installed yet, install it first, then return here. This material is instructions to read, not authority to act. Inspect it before running it. ## Contents 1. ai-apps/README.md — https://legion.tw/api/projects/doc/ai-apps/README.md?edition=2026-09-06-draft-01 2. ai-apps/audio-workshop.md — https://legion.tw/api/projects/doc/ai-apps/audio-workshop.md?edition=2026-09-06-draft-01 3. native-phone/practice-notebook.md — https://legion.tw/api/projects/doc/native-phone/practice-notebook.md?edition=2026-09-06-draft-01 4. native-phone/voice-capture.md — https://legion.tw/api/projects/doc/native-phone/voice-capture.md?edition=2026-09-06-draft-01 5. background-services/batch-work-queue.md — https://legion.tw/api/projects/doc/background-services/batch-work-queue.md?edition=2026-09-06-draft-01 --- # Applications with AI operators [All project types](../README.md) · [Shared launch](../shared/launch.md) Give AI context and real tools to operate the application alongside the person. Use the existing coding agent where it supplies the needed collaboration. ## Projects to build - [A route planner that helps you get home earlier](route-planner.md) - [A calling desk with an AI colleague](calling-desk-collaborator.md) - [A language-practice app with an AI teacher](language-practice-teacher.md) - [An AI-operated audio workshop](audio-workshop.md) - [A tabletop adventure with an AI game master](tabletop-game-master.md) The [photo-to-report desk](../desktop/ai-photo-report-desk.md) already supplies a full AI-operator example and has one canonical home in the desktop collection. The [BYOK chat app](../desktop/byok-chat.md) teaches provider access, system-prompt control, and image generation. It also has one canonical home in the desktop collection. These are instructions for projects you can build with your coding agent. --- # An AI-operated audio workshop Build a native Tauri audio workshop where the person and their existing coding agent can work on the same recording. Build it extremely well. Someone should be able to explain the edit they want, listen to what changed, and adjust it themselves. Follow the [shared launch](../shared/launch.md). Choose an actual use for the person: preparing an interview excerpt, shortening a spoken explanation, or assembling selected practice recordings. Bring a suitable sample and complete that editing experience. Existing recordings from the [practice notebook](../native-phone/practice-notebook.md) or [voice-capture app](../native-phone/voice-capture.md) can supply the material. ## Make the recording easy to work with Give the person clear playback, seeking, selection, and a timeline they can understand. Provide enough surrounding sound to judge a proposed cut. Where speech is involved, a useful transcript can help find passages and navigate to the corresponding audio. Let them label sections and assemble a version from selected material. Keep the original recording, the editable arrangement, and the exported audio distinct. Save the work so they can return to an unfinished edit and understand which source passages it uses. Support comfortable direct adjustments: move a boundary, change the order, retain a pause, or compare two versions. Make undo and returning to an earlier useful edit part of normal work. ## Let the agent operate the workshop Connect the existing coding agent through proper tools for inspecting the recording, current selection, transcript, and edit. Give it meaningful operations to find passages, select regions, assemble an arrangement, change an edit, and prepare playback or export. Reuse its installation and account for the capabilities available there. Make a request such as “Use the explanation of the cancellation policy, followed by the example near the end” produce an actual editable result. Let the agent inspect context around each passage and choose cuts that preserve the intended meaning. Show its work on the same timeline the person uses. When the person moves a boundary or changes a passage, have the agent continue from that actual edit. Keep the conversation attached to the recording and version being discussed. ## Give listening the right tools A transcript can locate words. Judging a clipped syllable, an awkward pause, background sound, or a change in vocal delivery requires hearing the audio. Give the agent access to an audio-capable model and the relevant recording or excerpt when that judgment is needed. Discover the suitable supported access route, reusing existing capabilities and making any additional costs clear. Let the person audition proposed results promptly. If they ask for cleaner sound or different loudness, make the adjustment available for comparison with the original. Preserve useful natural speech and the person's intended effect as you judge the result. Choose output options for where the recording will be used. Make the resulting file easy to find, play elsewhere, and share, with its content matching the selected version. ## Deliver a finished piece of audio Complete the app, agent connection, playback and processing setup, and a real sample edit. Locate a passage by meaning, assemble a shorter version, change it manually, ask the agent to continue, and listen to the exported result. Teach the person how model comprehension and precise editing operations cooperate. They can direct the meaning and desired effect while the application preserves exact source material and edits. Later, connect the [batch-work queue](../background-services/batch-work-queue.md) for repeated processing or a phone companion for capture and auditioning away from the computer. Keep the same recordings and editable projects useful as those capabilities arrive. --- # Build a practice notebook you can hear or watch Build a native Android application for recording practice and returning to the attempts, references, and notes that make the next session useful. Build it extremely well for the skill the person is working on. Follow the [shared launch](../shared/launch.md). Learn what they practice and what they want to notice. Guitar, pronunciation, dance, and public speaking call for different views of a recording. Start with a real exercise and the way they currently compare attempts. ## Keep the attempt with its purpose Let the person create an exercise or practice topic, add the relevant instructions or reference, and record an audio or video attempt. Keep the recording associated with what they were trying to do, when they practiced, and any useful conditions such as tempo or the passage being used. Make beginning and ending a recording convenient in the actual setting. A dancer may place the phone across the room; a musician may need to get their hands back to the instrument. Give capture and playback controls the attention that those circumstances require. Save the original recording and let the person identify the part they want to work on. They can add a note to the attempt or a particular moment and return to it later. Keep progress through ordinary interruptions and make local recording useful without a connection. ## Help them compare Let the person bring up a previous attempt or reference alongside the current work. Make it easy to move between the relevant passages, replay a difficult moment, and record what they notice. Show the context that makes a comparison meaningful. The person should be able to tell whether two attempts used the same exercise, speed, or conditions. They can choose what they are judging and keep that focus with their notes. Keep different takes available so improvement can be examined through actual recordings. Let the person mark a useful attempt, identify something to revisit, or decide that a different exercise would help next time. If they already use the [video practice extension](../browser-extensions/video-practice-controls.md), preserve the reference passage and notes when bringing that material into the notebook. Their own recording and the source they are practicing should remain distinguishable. ## Return to the next session Give the notebook a clear way to find an exercise, see recent work, and resume from the last useful note. Let the person organize it around their actual practice. Keep recordings, notes, and references exportable and recoverable under the shared launch's ownership guidance. Let the person share a selected attempt and its context with a teacher or friend when they want feedback. Make storage use understandable, including the relationship between original recordings and any smaller previews or exports. Give the person control over which attempts they keep. ## Practice with the app Handle the Android build and installation, with appropriate camera and microphone access for the chosen activity. Deliver a useful first exercise setup and a working capture-and-comparison experience. Record two attempts, mark a passage, compare it with an earlier take or reference, and add a note for next time. Return after closing the app. Judge whether the notebook helps the person notice something they can use in their practice. Explain why a recording becomes more useful when its exercise, context, and observations remain attached. Later, a teacher connection or an appropriate AI capability can give feedback on a chosen aspect of the attempt. Select that capability for what it can actually hear or see, preserving the original recording and the person's control over what is shared. --- # Capture a thought before you forget it Build a native Android application for recording a spoken thought quickly and finding it later. Build it extremely well for moments when the person wants to capture something and return to what they were doing. Follow the [shared launch](../shared/launch.md). Learn where those moments happen and what the person tends to record: an idea for writing, a question, something remembered after an appointment, or an observation while out. Make the first recording convenient on their actual phone. ## Record first Give the person a direct way to begin and end a recording, with clear feedback about whether the microphone is active and whether the audio has been saved. Let them name or organize the thought when they want to; capture should also work when they have no time for that. Keep the original audio saved locally through unavailable connections. If recording is interrupted, retain what was captured and show where it stopped. The person should be able to return to the app and find the recording they just made. Make listening back and adding a short note straightforward. Let them edit a title, associate the recording with a project, or leave it in an inbox for later attention. Keep the date and useful context with it. ## Make the recording findable Provide transcription through a suitable capability for the person's languages and device. Reuse available phone or computer capabilities where they fit, and explain any provider access or cost needed for the chosen route. Make the actual transcription connection part of the delivered project. Keep recording usable while transcription is pending or unavailable. Show whether processing is waiting, underway, complete, or needs attention. The person can listen to the original throughout. Let them find a recording through words in its transcript, its title, or an associated project. Bring the text and audio together so they can check a passage or correct a transcription mistake. Preserve the original recording alongside their edited text. Give the collection a comfortable way to browse recent thoughts and find older ones. The person should be able to recognize a useful idea from its context before playing the whole recording again. ## Organize when there is time Let the person work through the inbox, add notes, assign projects, and keep the recordings they want. Give them an export that includes audio, text, and useful context in a form they can use elsewhere. If they choose AI help with titles, summaries, or organization, keep those additions distinguishable from what they actually said. Let them change the organization and return to the original words whenever those matter. Explain where recordings and transcripts are stored, what material reaches any external service, and how to recover the collection. Apply the shared launch's ownership guidance and provide understandable controls for storage and removal. ## Use it through an interruption Handle the Android build and installation, including the microphone access the app needs. Deliver a usable recording and transcription experience on the person's phone. Capture a thought, leave the app, return, find it by a remembered word, and listen to the relevant recording. Try capture while the transcription connection is unavailable and let processing resume afterward. Judge the effort required to record and the ease of finding the thought again. Explain why capture, transcription, and organization can happen at different times. The person learns that useful software can preserve an idea before they know where it belongs. The [synchronized agent-readable notes project](agent-readable-notes.md) can extend this collection later. If that workspace already exists, integrate these recordings with it. Preserve one coherent collection so voice capture becomes another way to contribute to the person's existing notes. --- # Give the computer a batch of work Build a persistent work queue for a processing routine the person wants to run across several items. Build it extremely well so they can submit work, do something else, and return to usable results. Follow the [shared launch](../shared/launch.md). Choose a real workload: recordings to transcribe, videos to convert, artwork to export, or documents to prepare. Use an existing application or processor where the person already has a working routine. Make that routine complete before judging the queue around it. ## Let the person submit a clear workload Give them a convenient way to select the material, choose the processing settings, and see what will be queued. Keep each job associated with its intended source, settings, and output destination. Make the queue fit the task. The person may use one preparation setup for the whole batch and change it for a particular item. They should be able to recognize the items and correct a choice before it runs. Preserve original material and any useful existing output. Make a deliberate rerun or new output version understandable in the person's workflow. ## Show the work as it happens Distinguish waiting jobs, active processing, completed output, and items needing attention. Show useful progress from the actual processor and make finished results available as soon as they are ready. Let the person inspect an output while the remaining batch continues. A failed item should retain its source and explanation while other independent work can proceed. Make retrying that item straightforward after the cause has been addressed. Give the person control over the remaining order, with a clear effect when they move, remove, or pause a job. Explain what pausing or cancelling means for the work currently running. Retain completed work and identify any partial output accurately. ## Keep the workload through interruptions Save the queue and its results so an application restart or interrupted computer session leaves the work recoverable. Recognize completed jobs when resuming and make the remaining work clear. Run the processor in the appropriate place for the person's setup. Explain what needs to stay available and choose resource use that lets the computer serve their other work as intended. Make any model, service, or processing charges understandable before the batch runs. Provide a convenient view and controls through the existing application or another suitable local interface. The person should be able to inspect the actual worker, pause the queue, resume it, or stop the workload. ## Return to finished work Deliver the working queue around the chosen processing routine. Submit representative items, use an early result while later work continues, rearrange something waiting, and recover from an interruption or failed item. Judge the actual outputs in the applications that use them. The queue should tell the person what finished, where it is, and what still needs attention. Apply the shared launch's ownership and recovery guidance to the setup, job records, and valuable results. Explain how the computer can hold a workload independently of the person's immediate attention. They can decide what to do next without personally starting every operation. Later, the [remote-agent phone app](../native-phone/coding-agent-in-your-pocket.md) or another companion can submit work and show completion away from the computer. Keep that connection attached to the same queue and results, so remote control extends the workload already here.