Ask in plain language
Your AI agent sends one question about many conversations, such as "why do customers cancel?", and EnderGPT answers from those conversations. Follow-up questions go to the same chat, which uses the earlier messages as context.
Get one answer to why customers cancel or call again, drawn from last week's calls, with nothing exported by hand. Your team's own AI agent, built with Claude, ChatGPT or another AI, asks through Ender Turing's REST API. EnderGPT answers from the calls in that scope, and by default a research answer ends with a numbered list of its sources.
An administrator invites a user for your AI agent, ticks System/Technical User and gives it a role that may create EnderGPT chats, with access to analytics data, and that sees the conversations it will ask about. That role decides everything your AI agent can see. API access ↗
One public plain-text page tells an AI agent how to send its token, scope a question with filters, ask EnderGPT (Ender chat in the guide) and wait for the answer. Give it the address and a token of its own account; when the token expires, your AI agent asks you for a new one. Guide for AI agents ↗
Your AI agent asks in plain language, for example why customers cancel, scoped to last week's calls, or to one topic within them. It waits for the answer, which can take minutes, then asks follow-up questions in the same chat. Chat with EnderGPT ↗
Every item below is in the guide for AI agents or in the Help Center. The links on each card are the sources.
Your AI agent sends one question about many conversations, such as "why do customers cancel?", and EnderGPT answers from those conversations. Follow-up questions go to the same chat, which uses the earlier messages as context.
The question takes filters written the same way as on the Conversations screen, Ender Turing's list of conversations: a date range, a team or a contact-center agent, a topic, inbound or outbound, a length. Scope it to the calls that can answer the question.
By default, a research answer is a report in headed sections that ends with a numbered list of sources. A format the question asks for, such as five bullets or one table, takes priority.
Your AI agent counts the calls that match the same filters, including the calls a text search finds in what was said, comments, summaries or phone numbers, such as "cancel my subscription". The count is exact, with their total length: EnderGPT answers why, the count answers how many.
When a help desk or CRM ticket names a call, your AI agent searches for the ID the call came with, such as your phone system's call ID, and keeps the conversations whose stored ID equals it exactly. Which IDs your calls carry depends on how they reach Ender Turing, and one call recorded in parts can come back as several conversations. Then it gets the call's details, every quality score point by point, and its summaries where they are set up.
Your AI agent can send the same question every week, scoped to last week's calls. Or, with a role that may manage scheduled chats, it sets up a scheduled chat that runs every day, week or month and sends each result as a report to the people you choose. The request is in the full API reference, which opens only for an account that may also create conversations; otherwise your AI agent asks you for it.
What the API leaves out: Transcript text is not part of the published API: for what was said, your AI agent asks EnderGPT or gets a conversation's summaries. Which plans include API access is in the plan comparison.
EnderGPT answers from the conversations in the question's scope, and a research answer lists its sources by default. These steps, from the Help Center and the guide, help keep each answer tied to your calls.
Yes. Through Ender Turing's REST API, your AI agent, built with Claude, ChatGPT or another AI, sends EnderGPT a natural-language question, waits for the answer, which can take minutes, and gets one answer back; by default, a research answer ends with a numbered list of its sources. This needs a plan that includes API access, EnderGPT switched on for your organization, an account for your AI agent whose role may create chats with access to analytics data, and an AI agent that can send web requests with its own headers. The public guide for AI agents shows it how, follow-up questions included. Ender Turing is headless conversation analytics for contact centers, built for AI agents: an agent or another system works with it through an API and a public guide, without the web interface.
Through the API, your AI agent gets the answer's text; by default, a research answer ends with a numbered list of its sources, and your AI agent can list the conversations that match the same filters to see which calls were in scope. In the web app, each quote is tied to the conversation and transcript line it came from, each conversation link opens that conversation, and a quote that does not closely match the transcript carries a warning. Your AI agent's chats belong to its account, so a person reads them there with a role that has the All scope for chats, which an administrator adds to a role of your own.
Any AI agent that can send web requests with its own headers, such as one your team builds on Claude or ChatGPT. It sends a token of its account in one header and names itself in another, as the guide shows, and gets JSON back. If a firewall page comes back instead, Ender Turing support can add an exception for your AI agent. No AI agent yet? Ask the same question in the web app's Chat with EnderGPT.
No. Transcript text is not part of the published API. For what was said, your AI agent asks EnderGPT, which answers from the conversations, or gets a conversation's summaries where summaries are set up. If your integration needs raw transcripts, tell us the use case when you book a demo: the right approach depends on your deployment.
Only what its account may see. Every endpoint checks its own permission, and a role limited to one team, or to its own conversations, returns only those, as in the web interface. To ask EnderGPT, the role needs permission to create chats and access to analytics data.
A research answer can take minutes. Your AI agent checks back every few seconds until the answer is ready; a question sent before then is refused, so it waits before it follows up.
Yes. Your AI agent can send the same question each week, scoped to last week's calls. Or, with a role that may manage scheduled chats, it can set up a scheduled chat that runs every day, week or month; the schedule sets the dates, so its filters leave the date out. The request is in the full API reference, which opens only for an account that may also create conversations; otherwise your AI agent asks you for it. Each result goes as a report to the people you choose, or as a separate report about their own conversations to each active contact-center agent with an active linked login in the teams or agents you select.
Security, hosting regions, sub-processors and response times are documented on the Data Security page; SOC 2 Type II audited, the latest report available under NDA.
More jobs for your AI agent: BI tool, warehouse and CRM · Weekly quality review · ten platforms compared.
Book a demo and bring the question you want your AI agent to ask about your calls every week, such as why customers cancel or call again.