Otter.ai

Breadth across classes, and the connectors run in both directions rather than only ingesting. Conferencing platforms supply capture. A public API covers Jira for issue tracking, HubSpot and Salesforce for CRM, and proprietary systems, with webhooks on Enterprise. AI Chat Connectors talk to Google Drive in both directions, so documents come in and outputs go back out.

CRM is queryable from inside AI Chat, meaning a rep asks about a deal without leaving the conversation, and for sales specifically the platform extracts BANT fields, syncs the CRM and drafts follow-ups. An MCP server exposes the corpus to external models. Otter is additionally distributed as a native connector inside Glean.

Sourceotter.ai enterprise suite announcement and product blogread 2026-09-04

Three named agents doing multi-step work, and one of them runs an entire commercial motion. The Meeting Agent participates in a live call, answers questions from company-wide meeting data and completes tasks during the meeting rather than after it. Agentic AI Chat autonomously handles meeting tasks including scheduling and action item capture. The SDR Agent conducts a full product demo with a prospect end to end.

Enterprise adds unlimited custom AI workflows over the top. The SDR Agent stands out for its autonomy: it performs the sales conversation itself, live, to a stranger, rather than writing, scheduling or updating a record, and Otter runs it on its own website before selling it. That is why the lack of any documented evaluation or approval mechanism for it matters more than the same gap would elsewhere.

Sourceotter.ai agent suite launch and enterprise suite announcementread 2026-09-04

The whole positioning of the October 2025 enterprise suite is that the meeting archive is a corporate knowledge base rather than a set of transcripts, and the agents ground on it.

Agents draw on company-wide meeting data across more than a billion meetings processed, AI Chat queries Salesforce and HubSpot directly from inside the conversation so CRM state is part of the same context, document integrations connect Google Drive in both directions, and screenshot content is incorporated automatically so visual material shared in a meeting becomes retrievable context rather than being lost with the recording.

Transcription accuracy is strong in English and multilingual coverage is weaker, so the quality of the grounding corpus varies by language in a way the product does not surface.

Sourceotter.ai enterprise suite announcementread 2026-09-04

Control over the platform, and none over the agent that matters most. What is documented: advanced admin controls, access controls at user and team level, two-factor authentication, and customer-defined retention where an admin sets a duration governing every conversation in the workspace. The Sales Agent is advisory by design, coaching a human rep who decides what to say.

The two autonomous paths have no gate: the SDR Agent conducts live product demos with prospects entirely on its own, and Agentic AI Chat handles scheduling and action item capture autonomously, with no approval step, review queue, confidence threshold or confirmation documented for either.

That matters because of what this agent does: an unsupervised agent speaking to a prospective customer is high consequence autonomy, the output is a live commercial conversation rather than a draft someone reads later, and the only documented control is whether the agent is deployed at all.

Sourcehelp.otter.ai and otter.ai enterprise suite announcementread 2026-09-04

Attestations and access controls are both published first party. Otter holds SOC 2 Type II, publishing both the completion of the audit and receipt of the attestation report, and the help center restates the certification directly. HIPAA compliance was achieved in July 2025 through an independent assessment, with Business Associate Agreements available on the Enterprise plan.

For access control, Otter offers single sign-on and SCIM provisioning with enterprise domain capture, access controls at user and team level, two-factor authentication, and customer-defined retention where an admin sets a duration after which all workspace conversations delete automatically, with residual data removed on contract termination. Encryption and regular security audits documented alongside. HIPAA coverage and the BAA come only with Enterprise, so lower tiers are not covered for protected health information.

Sourceotter.ai HIPAA and SOC 2 announcements, help.otter.ai HIPAA article and the enterprise suite announcementread 2026-09-04

Every conversation is captured, searchable and replayable, and none of that records the agent. Advanced admin controls give organization-wide visibility, the archive makes any meeting traceable after the fact, and video replay on Enterprise means a manager can watch what happened rather than read a summary of it.

That is a record of the meeting, not of the system: no agent decision log, action history or reasoning trace is documented, so nothing shows what the Meeting Agent did when it completed a task during a call, what Agentic AI Chat scheduled on someone's behalf, or what the SDR Agent said to a prospect and on what basis. The last of those is the sharp one.

An unsupervised agent conducts a live demo to a stranger and the only artifact is whatever the platform happens to capture of the session; there is no documented trail a sales leader could audit if a claim were misstated.

Sourceotter.ai enterprise suite announcement and help.otter.airead 2026-09-04

The meeting archive is content, not memory. Every conversation, decision and insight persists in a central repository that the agents and AI Chat query, which is the customer's knowledge; customer-defined retention governs the archive as a whole. No memory the agents write and read back as their own state, with a scope, lifetime or delete path of its own, is documented.

Sourceotter.ai and help.otter.airead 2026-10-01

Multi-tenant cloud with no location control offered. What Otter documents is governance over data rather than placement of it: enterprise domain capture, customer-defined retention policies with an admin-set duration after which all workspace conversations delete automatically, secure deletion, and removal of residual data on contract termination. Those are real controls and they are lifecycle rather than location. No region selection, EU or other in region option, private cloud, VPC, single tenant tier, on premises route or customer owned storage is documented.

Sourceotter.ai enterprise suite announcement and help.otter.ai HIPAA articleread 2026-09-04

A named set of prebuilt agents, each doing its own job. The home page lists the Otter Meeting Agent, Sales Agent, Education Agent, Media Agent, Recruiting Agent and SDR Agent, the SDR Agent taking website visitors from a live demo to a booked meeting and the Sales Agent coaching live and syncing insights to the CRM; each stands without the others.

Sourceotter.ai and otter.ai/pricingread 2026-10-01

Three genuinely different trigger surfaces, including one that starts from outside the customer's organization. Scheduled: OtterPilot joins meetings from the calendar without anyone starting it. In-meeting and voice-activated: the Meeting Agent is addressed by voice during a live call and acts while the conversation is happening, which is a narrower window than before-or-after.

Inbound and external: the SDR Agent is triggered by a website visitor arriving at any hour, so the initiating party is a prospect rather than an employee, which is unusual. Enterprise adds webhooks for programmatic triggering. Channels out span the meeting itself, AI Chat, CRM writes and follow-up drafts.

Sourceotter.ai agent suite launch and enterprise suite announcementread 2026-09-04

The customer does not choose the model that powers Otter's own agents, and no provider is named. The MCP server connecting Otter data to external AI models, including Claude, means the customer's assistant reads the vendor: the customer chooses the interface, not the engine. No published page names a model provider behind the Meeting Agent, Sales Agent or SDR Agent, or offers selection, routing or a bring your own key option for them.

Sourceotter.ai enterprise suite announcement and help.otter.airead 2026-09-04

Two external surfaces and a two-way connector set. A public API supports custom integrations with named platforms including Jira, HubSpot and Salesforce as well as proprietary systems, with webhook access on Enterprise and unlimited custom AI workflows. An MCP server connects Otter data to external AI models including Claude, so a customer's own assistant reads the meeting corpus directly. No SDK or client library is published, and API and webhook access come only with the Enterprise tier. Otter is also available as a native connector inside Glean.

Sourceotter.ai enterprise suite announcementread 2026-09-04

Nothing lets a customer test what the agents produce, before or after. No evaluation harness, scored test set, benchmark, regression check, sandbox or dry run is published, and no scoring of agent output quality is documented.

The gap is sharpest for the SDR Agent, which conducts unsupervised live demos to prospects, so its output is a commercial conversation with a stranger, delivered in real time, with no supplied way for the customer to have measured it first and no way to test a change before it goes live on their website. The same applies to Agentic AI Chat scheduling on someone's behalf.

Where a lower-consequence product can reasonably ship without an evaluation surface, this one carries the customer's brand into a first conversation with a buyer.

Sourceotter.ai agent suite launch and enterprise suite announcementread 2026-09-04

Two features sound like computer use and are not. The SDR Agent is described as multimodal and conducts live product demos, and Agentic AI Chat incorporates screenshot content. The demos run on Otter's own website, demonstrating Otter's own product, so whatever interface the agent drives is one the vendor controls and has a programmatic route into.

Computer use means driving software the vendor does not control and cannot reach through an interface; a product's own screens are a different thing. Screenshot ingestion is reading an image, not operating the application that produced it. Everything else moves through conferencing platforms as a participant, the public API, webhooks and the MCP server.

Sourceotter.ai agent suite launch and enterprise suite announcementread 2026-09-04