LiveCopilot

Listening · Remote Meeting · Them ready

Question ready
Them

If you could ship only one feature first, how would you prioritize it?

If you could ship only one feature first, how would you prioritize it?

Suggested answer

I would weigh user value, development cost, and whether this feature enables what comes next. If it serves a frequent need and unlocks future capabilities, I would ship it first.

Evidence & notes

The product strategy prioritizes frequent needs; the Roadmap puts core capabilities before later expansion. [S1][S2]

Sources · retrieved local evidence

[S1] Product Strategy.md · Priorities

Compare user reach, how often the problem occurs, and development cost. Give extra weight to features that enable later work.

[S2] Roadmap.docx · Q3

Core capabilities come before later expansion.

Meeting⌥H

Based on the real overlay. Scripted materials and answers; no recording or AI calls. Surrounding documents illustrate the process.

The conversation continues.Your next thought is ready.

Hear the conversation. Bring in your own material. Find the words to respond, right on your Mac.

Download for macOSGitHub

macOS 14+ · Intel & Apple Silicon · Free & open source

Explore how LiveCopilot works

From hearing the question to finding your words.

Connect the conversation to your material. Put your preparation to work.

  1. Hear

    Hear

    Turn microphone or system audio into live text.

    If you could ship only one feature first, how would you prioritize it?

    Question recognized
  2. Retrieve

    Retrieve

    Find the relevant passages in your own documents.

    Product Strategy.md User value · Frequent needsRoadmap.docx Core capability · DependenciesInterview Notes.pdf Build cost · User feedback3 relevant passages found
  3. Think

    Organize

    Bring the question, context, and references together.

    QuestionRecent conversationReferences
    Organizing answer… Extracting: user value · cost · dependencies
  4. Answer

    Answer

    Get a natural answer you can put into words.

    If you could ship only one feature first, how would you prioritize it?

    Suggested answer

    I would weigh user value, development cost, and whether this feature enables what comes next. If it serves a frequent need and unlocks future capabilities, I would ship it first.

    Evidence & notes

    The strategy highlights frequent user needs; the Roadmap puts core capabilities ahead of later expansion. [S1][S2]

    Sources · retrieved local evidence

    [S1] Product Strategy.md · Priorities

    Compare user reach, how often the problem occurs, and development cost. Give extra weight to features that enable later work.

    [S2] Roadmap.docx · Q3

    Core capabilities come before later expansion.

Present when you need it. Space when you don’t.

Follow the conversation

Follow live captions and keep the context of the question.

Overlay detail · Sample content

Them

If you could ship only one feature first, how would you prioritize it?

Your material, within reach

Bring PDF, DOCX, Markdown, or TXT files. Ground the answer in your work.

Retrieval illustration · Outside the app

PDFResearch.pdf
DOCXRoadmap.docx
MDStrategy.md✓ 3 relevant passages found
+ TXT

Tuck away. Return quietly.

The window fades away to leave room for your conversation. Move to the screen’s right edge to reveal it, or pin it.

LiveCopilot

Ready — type a question or start listening

Question ready
Them

If you could ship only one feature first, how would you prioritize it?

If you could ship only one feature first, how would you prioritize it?

Ask a question below, or press ⌃⌥Space to generate an answer from the conversation.

Suggested answer

I would weigh user value, development cost, and whether this feature enables what comes next. If it serves a frequent need and unlocks future capabilities, I would ship it first.

Evidence & notes

The product strategy prioritizes frequent needs; the Roadmap puts core capabilities before later expansion. [S1][S2]

Sources · retrieved local evidence

[S1] Product Strategy.md · Priorities

Compare user reach, how often the problem occurs, and development cost. Give extra weight to features that enable later work.

[S2] Roadmap.docx · Q3

Core capabilities come before later expansion.

Meeting⌥H
Docked at the right; no persistent handle when hidden

Words you can actually say

The overlay pairs a speakable answer with notes and expandable source passages.

Overlay detail · Sample content

If you could ship only one feature first, how would you prioritize it?

Suggested answer

I would weigh user value, development cost, and whether this feature enables what comes next. If it serves a frequent need and unlocks future capabilities, I would ship it first.

Evidence & notes

The product strategy prioritizes frequent needs; the Roadmap puts core capabilities before later expansion. [S1][S2]

Sources · retrieved local evidence

[S1] Product Strategy.md · Priorities

Compare user reach, how often the problem occurs, and development cost. Give extra weight to features that enable later work.

[S2] Roadmap.docx · Q3

Core capabilities come before later expansion.

Different conversations. The same prepared you.

Question

If you could ship only one feature first, how would you prioritize it?

Product Strategy.mdRoadmap.docx
Retrieved passage

Compare user reach, how often the problem occurs, and development cost. Give extra weight to features that enable later work.

With conversation context · Live suggestion

If you could ship only one feature first, how would you prioritize it?

Suggested answer

I would weigh user value, development cost, and whether this feature enables what comes next. If it serves a frequent need and unlocks future capabilities, I would ship it first.

Evidence & notes

The product strategy prioritizes frequent needs; the Roadmap puts core capabilities before later expansion. [S1][S2]

Sources · retrieved local evidence

[S1] Product Strategy.md · Priorities

Compare user reach, how often the problem occurs, and development cost. Give extra weight to features that enable later work.

[S2] Roadmap.docx · Q3

Core capabilities come before later expansion.

Based on the real overlay. Scripted materials and answers; no recording or AI calls. Surrounding documents illustrate the process.

Hear. Retrieve. Respond. Choose a service for each.

See what stays local and what reaches a provider, step by step.

Hear

Local / Cloud
Apple · Paraformer · GPT-Live-1

Turn microphone or system audio into text.

Retrieve

Local / Cloud
BGE-M3 · OpenAI

Choose local or cloud embeddings; search stays on the Mac.

Respond

Cloud
OpenAI · DeepSeek · Qwen · GLM · Kimi

Send the question and relevant passages to your provider for an answer.

Compare pricing, limitations, and compatibility

How much stays local? You choose.

Choose speech recognition, document indexing, and answer services independently.

Shown with local speech and embedding models. Choosing cloud speech or embeddings also sends the corresponding audio or text to that service. Answer generation sends relevant context to your selected analysis service.

Compare pricing, limitations, and compatibility
Your Mac Local search
Microphone / system audioLocal transcriptionApple / ParaformerTranscript
Your documentsLocal embeddingsBGE-M3Local search
Question + relevant conversation + retrieved passages
Your chosen AI provider
Your answer

Your conversation
is yours.

Your words. Your choice.

No LiveCopilot account

No LiveCopilot cloud

Knowledge base can stay local

API keys in macOS Keychain

Your MacRelevant contextYour AI provider

Relevant context goes directly to your chosen AI provider, without a LiveCopilot server. The provider’s own data policies still apply.

Read the full privacy notes

Your next conversation. A little more composed.

↓ Download for macOS

Universal · Free & open source macOS 14+ · Intel & Apple Silicon

New in 2.1: guided setup and in-app updates. Check for updates in Settings → About, with optional automatic checks. You approve downloads and installation. Version 2.0 and earlier need this first upgrade installed manually.

Installation and first-time setup
  1. Quit the previous version and move LiveCopilot.app into Applications. Keep its installation location consistent.
  2. Follow the six-step welcome guide to choose your language, conversation mode, models, answer service, and audio permissions.
  3. Choose an analysis service, save your API key, and optionally import your own material.
  4. Start listening and grant the required audio permissions, or just type a question. Option + H shows the overlay.

Version 2.1.0 is Developer ID signed and Apple notarized, with tickets stapled to both the app and DMG. Upgrading from an older ad-hoc build may require renewed recording or Keychain permission. Apple installation guidance

Compare pricing, limitations, and compatibility

Choose a service for each part of the conversation.

Configure speech, document embeddings, and answers independently. Choose Apple or Paraformer plus BGE-M3 to keep audio and indexing on your Mac, then use your preferred analysis model for answers.

Listen to the conversation

All three options stream captions. Previews can be revised; committed transcripts enter the answer context.

Apple Speech

On-device · no API fee

Uses macOS SpeechAnalyzer / SpeechTranscriber. Apple manages the models, with offline recognition after download and less model setup to manage yourself.

Requires macOS 26+ and supported hardware. Select Mandarin or English before listening; this route does not switch languages automatically. Performance depends on your Mac and workload.

Paraformer

On-device · no API fee · about 238 MB

The bundled bilingual streaming model updates captions while you speak and supports Chinese/English conversation. Runs on macOS 14+ without Python or a separate inference server.

Requires a model download and local compute. English terminology, accents, and noise can cause errors; compare with your actual audio. Version 1.4 removes the SenseVoiceSmall sentence-recognition option.

GPT-Live-1

Cloud · $0.05 / minute / session

Adds live semantic decisions and delegation alongside streaming transcription. Choose Chinese-first, English-first, mixed, or any-language guidance; these hints do not lock transcription to a language.

Requires internet and sends audio to OpenAI. System audio plus microphone uses two sessions, about $0.10/minute; connected silence also counts. Analysis is billed separately. LiveCopilot does not play AI speech. Official pricing and details

Find your own material

Embeddings help find relevant passages; they do not write answers. The retrieval database stays on your Mac.

BGE-M3

On-device · no API fee · about 635 MB

Embeds documents and queries locally. Indexing and retrieval work offline after download, so document text need not be uploaded for indexing.

Initial loading and batch indexing use local memory and compute. Speed depends on the device. Re-index after switching embedding models; keyword search still works with older indexes.

OpenAI Embeddings

Cloud · Small $0.02 / 1M input tokens

No local embedding model to download and less local inference work. Small costs less; Large is $0.13 per 1M input tokens. Both document indexing and query embedding incur usage charges.

Extracted document text and retrieval queries go to OpenAI, requiring a connection and key. A larger model does not guarantee better retrieval on your material; compare using your own questions. Official pricing and details

Put the answer into words

Receives the question, relevant conversation, and retrieved passages; returns a spoken-style reply, recap, or follow-up.

These cloud services work with any speech or embedding option. More thinking usually means a longer wait and potentially more tokens. For live use, try optional thinking off or a low effort level, then compare quality on your own questions.

OpenAI

Cloud · billed per token

Reuse the OpenAI live-service key or supply a separate one. Streams answers through Responses with reasoning-effort controls for the selected model.

Analysis and Live are separate charges, even when they share a key. Check account access and pricing before changing models. Official pricing and details

DeepSeek

Cloud · billed per token

An existing dedicated integration that pairs with local speech and embeddings. Select an available Flash or Pro model and thinking effort; analysis is charged when you request it.

Pricing varies by model, cached input, and peak/off-peak periods. Thinking tokens count as output. Lower effort can reduce waiting, without guaranteeing lower cost or higher accuracy on every question. Official pricing and details

Qwen

Cloud · Alibaba Cloud Model Studio

Starts with Qwen Plus, streaming answers, and optional thinking. The model name is editable, useful if you already use Model Studio and want to choose models there.

Keys are region-specific; Beijing and Singapore credentials are not interchangeable. Workspace endpoints are supported. Model, region, context tier, and thinking mode affect pricing. Official pricing and details

GLM

Cloud · Zhipu general API

Starts with GLM-5.2, with an editable model and thinking mode. Use a matching Zhipu or international Z.AI endpoint to bring your account into the same answer workflow.

Availability and pricing can differ between domestic and international platforms. Coding Plan subscriptions and endpoints are not general API access; use an appropriate API key and balance. Official pricing and details

Kimi

Cloud · Moonshot API

Starts with Kimi K2.6, streaming output, and a thinking toggle. Use it to organize answers from conversation and retrieved passages, then compare with the other analysis options.

China and international accounts use different endpoints, keys, and billing. Thinking can delay the first answer text. The app displays final-answer content, not internal reasoning. Official pricing and details

Other services can use the OpenAI-compatible option with an HTTPS address, path, and model. Compatibility means streamed Chat Completions, not universal support for every provider or key. Qwen, GLM, and Kimi have protocol and mock-test coverage; verify real account access with your own key.

Pricing references checked September 16, 2026; the provider’s bill is authoritative. Local models still use storage, memory, and compute, with internet needed for the initial download. With both speech and embeddings local, answer generation still sends relevant text to your analysis service; a separate provider does not require an OpenAI key.

A small window that grows with the thought

Stays compact when empty and grows with text. Adjust transcript and answer sizes separately; choose glass, soft frosted, or white backgrounds, then pin the window or reveal it from the right screen edge.

A fresh start for the next conversation

One refresh clears the current transcript, answer, and context. Active listening continues as a fresh session. Your knowledge base and saved history stay intact.