AI Discussion Let multiple AIs debate a question — and listen to the answer. Record your meeting and turn it into minutes automatically. Talk by voice, and finish your documents, too. AI Discussion 3.2 unites multi-AI discussion, meeting transcription, real-time voice conversation, and document creation in one buy-once desktop "AI collaboration suite." Powered by top AI models (Claude, ChatGPT, Gemini, DeepSeek, Kimi, and more), it lets you hear a discussion aloud, record a meeting and transcribe it to a verbatim report, talk out loud, and finish documents through multi-AI cross-checking. ■ No monthly subscription — start free, then pay only for what you use No fixed monthly fee — $0 in months you don't use it. The providers' APIs are pay-as-you-go, with no surcharge, tickets, or handling fees: you pay each provider directly with your own key, no middleman markup. Start free: Google Gemini and OpenRouter offer free tiers, and DeepSeek gives new accounts a free allowance (check each provider's terms). Even for serious use, OpenAI, Anthropic (Claude), and Kimi (Moonshot) start from small prepaid credits, and local Ollama is completely free. For everyday use, lean on fast, low-cost models; when it matters, switch to the latest top-tier ones (OpenAI GPT-6-astra, Claude Fable 5.1, Gemini 3.8 Flash, Kimi K3, and more). You choose both price and model per task. [The features below follow the same order as the app screen: the ? Dedicated row → ? Voice → ? Text] ■ Choose your model from real data — ? Benchmark Run identical, fair tasks across multiple AI models, objectively measuring Pass Rate (%), Generation Speed (chars/sec), Elapsed Time, and Token/Character counts — so you pick the best model for each job from real data, not guesswork. Add custom tasks or run a generate-only comparison, and watch progress, outputs, and pass/fail reasoning in real time on a dedicated Run View. ■ Interactive real-time voice — ? Live Chat (Voice selection) Real-time voice discussion on a dedicated page, fully independent of the other modes. Have Gemini Live and GPT Live talk simultaneously — you and up to two live AIs talk out loud. You speak (push-to-talk) or type, an AI replies aloud, and the floor returns to you. Pick a voice for each participant, and give each AI a nickname so that calling it by name makes only that AI answer. Gemini can also search the web mid-conversation (search queries shown on screen). ? The real edge is Context: pre-load the situation, role, and conditions via ? File, ? URL, or ? Paste. Use it as a pre-meeting sounding board (rehearse tough Q&A and sharpen your points), or to practice interviews, one-on-ones, sales calls, negotiations, and presentations. Prime the AI persona with exact behavior — a "tough interviewer," a "skeptical executive" — for realistic, high-stakes rehearsal. This kind of up-front Context conditioning isn't available in the standard phone Gemini / ChatGPT voice assistants. ■ Learn a language by speaking — ? Live Learning (Language Practice) Practice speaking through a real-time voice conversation with a native-teacher AI in your target language. The teacher adapts to your level (Beginner: slow and simple, Intermediate: everyday conversation, Advanced: native speed), and you can choose a situation (Daily Conversation / Travel / Cafe / Business / Free Talk). You can also add a bilingual "? Voice Assistant" that gently offers a hint only when you call out "Assistant." Supported languages: English, Chinese (Simplified / Traditional), Korean, Spanish, French, German, Portuguese, Japanese (9 languages). ■ Live on your phone — ? Live Companion (Android) Use Live — both Live Chat and Language Practice — from your phone. Show the QR code in the PC app and scan it with a phone on the same Wi-Fi (no install needed): your phone becomes a wireless mic and screen, to talk with the AIs or practice speaking hands-free, away from your desk. The settings you chose in the PC app carry straight over, across all 9 languages. Your API keys stay on the PC — nothing is handed to the phone. ※ The phone companion is available on Android devices only (iPhone / iOS is not supported). ■ Record your meeting and transcribe it accurately — ?️ Meeting (Report) (dedicated dashboard) A dedicated page for recording and transcribing meetings, opened from the "? Dedicated" row at the top. It records your system audio (remote participants from Zoom, Teams, Meet, or a browser) together with your own microphone, and transcribes it accurately with OpenAI Whisper into a verbatim record (Report). It supports 9 spoken languages — English, Japanese, Chinese (Simplified/Traditional), Korean, Spanish, French, German, and Portuguese — plus automatic language detection, so even multilingual meetings turn straight into a report. Recordings are saved so you can manage them later and re-transcribe on demand, and the Report exports to Markdown (.md). Load that Report into the ? Minutes mode and several top AIs cross-check it through Draft → Review → Final into structured minutes — summary, decisions, and TODO — ready for Word (.docx). ■ Sit in on your meeting, too — ?️ Meeting (Preview) (experimental) There is also an experimental mode where an AI can join an online meeting, listen to what's said, and respond. It is currently a prototype (in development), so behavior may be unstable depending on your environment, and its specification may change substantially or be removed in a future update. ■ Hear the discussion — ? Voice Discussion Each AI's turn is read aloud in turn, with a different voice per participant so you can tell who's speaking by ear. ■ Read and refine at your own pace — ? Text (5 modes) Run everything in text alone, too. Choose from five modes (in screen order): ? Discussion — Deepen brainstorming and debate from many angles. ? Plan — Turn ideas into a concrete plan: Propose → Critique → Improve → Action Plan. ? Doc — Produce polished documents from your attachments in Draft → Review → Final. Create from scratch, or review and revise a document you already have. Export to Word (.docx), Markdown (.md), or Plain Text (.txt). ? Minutes — Build structured minutes from your agenda or meeting notes, with a summary, decisions, TODO (task / owner / due date), and pending items. ⚙️ Custom — Freely design the number of phases, per-phase instructions, and which AI handles each — multi-step reviews for contracts and legal, finance, code and security, or fact-checking, cross-checked by AIs from different companies. Ready-to-paste recipes are included in the README. ■ Rich context loading — Files (.docx, .pdf, .md...), URLs & text paste Load reference material by file upload (Word (.docx), PDF, Markdown, text, CSV, JSON, YAML, etc.), text paste, or web-page URLs that fetch article text automatically — for discussions grounded in the latest information. ■ Strong privacy — data goes directly from your PC to each AI (never through our servers) What you type or say never passes through our servers: it goes encrypted, straight from your PC to each AI's official API, and we retain nothing. With the paid APIs of OpenAI, Anthropic (Claude), and Google (Gemini), your input is not used to train their models (per each provider's API policy) — dependable for confidential meetings and business documents. Note: on free tiers (e.g., Gemini's), data may be used for product improvement. For sensitive data, use a paid API or local Ollama — nothing leaves your machine. ■ Encrypt your API keys with a password — opt-in (optional) For stronger protection, set a master password to encrypt your API keys (PBKDF2 + AES-256). Skip it and the app starts with no password as before — fully backward compatible. Forget it, and a reset safely wipes the keys and re-initializes. ■ Supported AI providers OpenAI ／ Anthropic Claude ／ Google Gemini ／ DeepSeek ／ Kimi (Moonshot) ／ OpenRouter models ／ Ollama (local LLM) — cloud or local, your choice of AIs. ■ When it helps When you want an accurate record of a meeting kept as structured minutes afterward. When you want the AIs' input without reading the screen closely. When you need to review proposals, contracts, figures, or code from many angles before submission. For daily language practice, research, development, or business planning. ■ Full guide in the included README — every feature, plus ready-to-use recipes Open the built-in README (PDF, English) from within the app. It explains every mode and feature in screen order, and for ⚙️ Custom mode it includes ready-to-paste phase recipes for each review type. ■ Before you begin This is a desktop app for Windows PCs; it does not run on smartphones or tablets. API usage is billed separately by each provider. This product does not include the API fees of OpenAI, Anthropic, Google, DeepSeek, Kimi, OpenRouter, or others. Reading aloud, speech recognition, Live, Meeting (Report), and Language Practice also incur API fees from the corresponding services. Meeting (Report) uses OpenAI Whisper for transcription (an OpenAI API key is required). Meeting (Report) records your Windows system audio (WASAPI loopback), so remote participants' voices are captured even without a microphone; a microphone is needed to include your own speech. Voice input for Voice, Live, and Language Practice requires a microphone, and we recommend a headset for Live and Language Practice. Obtaining and managing your API keys is your own responsibility. AI-generated responses may contain inaccuracies, omissions, or outdated information — verify anything important yourself. Accuracy, completeness, and fitness are not guaranteed. For professional judgments in law, finance, medicine, and similar fields, always confirm with a qualified professional. ■ Support For questions or to report a problem, please contact so.tech.ken@gmail.com. Thank you for using AI Discussion. Many AIs for a single question — meetings into records — and finishing your documents, too.