Earshot

You just got off a customer call and three things they said were worth keeping. They go into a doc, the doc goes into a folder, and three weeks later the roadmap discussion runs on memory anyway. Someone asks "what are customers actually saying about pricing?" and the honest answer is that nobody knows where to look.

Earshot is an insight database your AI keeps for you. Paste the transcript into Claude (or Cursor, or any MCP client) and say "file this in Earshot." The vault stores the raw material exactly as you gave it, pulls out the individual findings, and tags each one with a theme, a sentiment, a one-sentence summary, and the customer's exact words. There is nothing else to set up. Research tools in this category sell you call-recorder integrations and import pipelines; Earshot skips all of it because the agent you already use is the pipeline.

Then you ask it things. "What's blocking activation?" comes back as a short answer with numbered citations, and behind every citation is a real quote from a real customer, linked to the full transcript it came from. The dashboard keeps score over time: themes accumulate counts, sentiment splits, and last-heard dates, so the third time someone mentions exports you see a pattern instead of a coincidence.

Bring your own AI key (Anthropic, OpenAI, or Google) for extraction and answers. An OpenAI key also turns on semantic search; without one, Earshot quietly falls back to keyword search and everything keeps working. An optional Monday digest emails you what the vault heard that week, and stays silent when there's nothing new.

Your sources, insights, and themes live in your own project's Postgres. Copy & Launch, connect your agent, and file your first call — the whole setup is pasting one URL.