# Productivity AI Assistant: Pocket, Now on AgentPMT > Pocket is now available as a connector on the AgentPMT marketplace, giving agents keyword and semantic search across a user's recordings, transcripts, and AI summaries, plus audio upload, temporary download links, and tag browsing. It turns a recording library into a queryable source that automations can act on, routing what was said into CRM notes, tasks, follow-ups, and documentation. Content type: article Source URL: https://www.agentpmt.com/articles/productivity-ai-assistant-pocket-now-on-agentpmt Markdown URL: https://www.agentpmt.com/articles/productivity-ai-assistant-pocket-now-on-agentpmt?format=agent-md Updated: 2026-08-03T21:16:21.795Z Author: Waffles Tags: MCP, AI Agents In Business, Product Releases, Product Specific --- # Pocket Recordings Now Run Inside Your Automations Every conversation you recorded last quarter is sitting in an app your other tools cannot reach. Capture got solved years ago. Getting a recorded decision into the CRM, the task tracker, and the follow-up email before anyone forgets it is still manual labor. [Pocket](https://www.agentpmt.com/marketplace/pocket-recordings-transcripts) is live on the AgentPMT marketplace as a connector, and it hands a productivity AI assistant direct access to your recording library: the audio, the transcripts, and the AI summaries already attached to them. Agents find it and call it through AgentPMT's dynamic MCP server, so any assistant wired into your account can pull up what was said without you exporting a single file. [Automated business search](https://www.agentpmt.com/articles/automated-business-search-for-ai-agents-on-agentpmt) is the capability to sit with. You can query the library by keyword or by meaning, which counts for a lot, because people remember the shape of a conversation rather than its exact phrasing. Semantic retrieval across your own recordings turns a vague "that call where the vendor pushed back on the timeline" into the actual transcript. From there an agent can pull a single recording with its full transcript and summarization data, browse by date or tag, and read back the tags in your library to learn how you organize things. Audio moves in both directions. An agent can mint a temporary download link, save a recording's audio into File Manager for something else to chew on, or upload an audio file you already have so Pocket can transcribe it. Old archives, phone memos, and files from other devices all land in the same searchable library. Billing is pay-per-use with no required subscription, so an automation that touches a handful of recordings a week stays proportional to what it actually does. The teams who already record everything are easy to name: field service techs walking a machine, advisors documenting client meetings for compliance, agency leads running back-to-back discovery calls, clinicians dictating between appointments. All of them capture faithfully, then lose hours re-listening. The composition that fits Pocket most directly already runs on AgentPMT with [a different recorder](https://www.agentpmt.com/articles/productivity-ai-assistant-plaud-now-on-agentpmt) attached. [Narrated Walkthrough to a Numbered SOP Document](https://www.agentpmt.com/agent-workflow-skills/narrated-walkthrough-to-a-numbered-sop-document) takes someone talking through a job out loud, walking the equipment and saying each step number as they go. It pulls the transcript of every new recording, judges whether the audio is genuinely a procedure, and writes a numbered standard operating procedure into a Google Doc with tools, safety notes, and a flagged list of anything the narration left ambiguous. Each run gets logged to a sheet so no recording is processed twice. Swap the capture device for Pocket and the shape holds: tribal knowledge in a technician's head becomes documentation, without anyone sitting down to write it. Client work composes the same way. A webhook fires the moment a recording lands, an agent parses the transcript for commitments and dates, writes the notes onto the right contact, opens the tasks, and drafts the follow-up. Prefer a digest? Put it on a nightly schedule that sweeps the day's recordings instead. Assemble that pipeline conversationally with the [Workflow Builder Agent](https://www.agentpmt.com/agent-workflow-skills/workflow-builder-agent), which walks you from an idea to a published skill across a catalog of well over a hundred tools, or dig through the [full workflow catalog](https://www.agentpmt.com/marketplace?types=workflow) for pieces to bolt on. Drive any of it from Hermes, OpenClaw, Codex, Claude Code, or ChatGPT, or let it run fully autonomously with no local agent setup at all. Wearable recorders have become standard equipment for people whose working day is mostly conversations, and their owners have been asking for the same thing for a while now: a way to plug the device into the assistants and systems they already live in, the chat tool they think in, the CRM their pipeline sits in, the notes app that holds everything else. Recorded audio is one of the last large unstructured data sources inside most companies, and business workflow automation has mostly routed around it, because the content stayed locked behind a player UI. Treating a recording library as a queryable memory store, something an agent reads from the way it reads any other system of record, is what separates a gadget from real business automation. That is the difference between owning a recorder and automating work that starts as a conversation. Pocket is on the marketplace now. Wire it into an assistant, ask it what you committed to in that meeting last Tuesday, and let the follow-ups write themselves.