Parley

Parley does four things: record, transcribe, analyze, search. It captures both sides of a call, gives you a speaker-labelled transcript live, writes a debrief when the meeting ends, and makes every word you have ever recorded searchable.

Then it opens all of it over a local MCP server — 54 tools — so Claude, Claude Code, Cursor or any other MCP client can read, search, file and re-analyze your whole meeting history without a copy-paste in sight.

• 🎙️ Both sides, live — your mic and the meeting's system audio, transcribed and diarized as you talk.
• 📼 A debrief that stays — commitments, action items, findings on a timeline, and a delivery scorecard. Generated once, saved with the recording.
• 🔎 Searchable forever — full-text across every transcript line and every conclusion the analysis drew.
• 🔌 Open by protocol — anything the app can do to your library, an MCP client can do too, including owning the analysis outright.

Local-first, bring your own keys. Audio and transcripts go directly to the STT and LLM providers you configure (Claude, OpenAI, Gemini, Soniox, Deepgram, …). No Pathors proxy, no telemetry, everything stored on your machine.

macOS and Windows, from one codebase. Both record both sides of a call: macOS captures the other side through a Core Audio process tap, Windows through WASAPI loopback of the default output device (no permission needed).

	macOS	Windows
Records the other side of the call	Yes — Core Audio process tap	Yes — WASAPI loopback of the default output device
Records your microphone	Yes	Yes
Voice typing	Yes (Option+Space)	Yes (Ctrl+Alt+Space)
Transcription, diarization and analysis	Yes	Yes
Sign-in and cloud sync	Yes	Yes
Closing the window	Hides it; Parley stays in the Dock	Hides it; Parley stays in the notification area (tray)

Closing the window never stops voice typing: Parley keeps running so the shortcut still works (an active meeting is ended and saved first). Bring the window back from the Dock (macOS) or by clicking the tray icon (Windows); quit with ⌘Q on macOS, or Quit Parley in the tray icon's right-click menu on Windows.

There are also companion apps for iPhone and Android that record in-person meetings and sync to the same account.

---

📥 Install

macOS — download the .dmg from the Releases page or directly, open it, and drag Parley into Applications. Builds are signed and notarized — no Gatekeeper hoops.

Windows — download Parley-setup.exe and run it. The installer is not code-signed, so SmartScreen shows an "unknown publisher" warning: click More info, then Run anyway.

Then paste your API keys in Settings: one STT provider for transcription, one LLM for analysis and Ask. Or sign in and use the hosted providers with no key to manage.

---

🎙️ During the call

Starting a meeting is one button. Parley captures both sides — your mic and the meeting's system audio — and transcribes them live, diarized as me / them, with editable speaker names.

Describe the meeting in its context field — who is in the room, what is at stake, what you want the analysis to watch for — and that description goes verbatim into every analysis prompt for the recording. It is also the field an MCP client writes to when it wants to re-aim the analysis.

Beside the transcript sits the coach feed: evaluation alerts from your own templates (negotiation risk, qualification gaps, red flags, or whatever rubric you wrote), each with a drill-down into how to reply, plus an agenda checklist that ticks itself off as the conversation covers it. Ask anything from the input bar and get answers grounded in the conversation so far.

---

📼 After the call

Stopping a meeting lands on its debrief. Any recording — just finished, pulled from history, dragged in as an audio file, or imported as a .txt transcript — opens in two views:

• Report — one scroll: the debrief with clickable timestamps, both sides' commitments, action items, and your delivery scorecard (measured pace, talk share, filler sounds).
• Replay — the full player: scrub to any moment and re-run the analysis as of that point, click through findings on the timeline, and ask anything about the call from a drawer that follows you across tabs.

Everything generates once and saves with the recording — reopen it a month later and the whole report loads instantly, no extra LLM calls. LLM speaker re-attribution fixes diarization drift by conversational context.

---

🗂️ Between calls

One customer, one folder. Every recording is filed in a folder — moved from the library, from the replay titlebar, or by an MCP client — so the calls with one customer sit together instead of scattering across a flat list. Sign in and a folder can live on the shared organization side instead of your personal one.

search_meetings runs full-text across everything you have recorded: every transcript line (each hit carrying a seek target) plus every brief, finding, action item and meeting context. Scope it to one folder to ask a question about one customer.

---

🔌 The MCP server

Parley runs a local Model Context Protocol server. Point your MCP client at it — the app has to be open — and it gets the live meeting, every saved recording, and the whole filing system:

claude mcp add --transport http parley http://127.0.0.1:3011/mcp

The exact endpoint, a ready-to-paste command, and a live view of what your client is doing are in Settings → MCP Server. 54 tools, in six groups:

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗