Conversations

Conversations are the organizing idea in Auron. They are how people use agents, and they are the evidence layer that summaries, signals, memory updates and actions are all derived from. An agent can be configured well, but a conversation is where the configuration meets a real situation: it decides what the agent sees, what it produces, and what it is allowed to start.

There are two ways to have one, voice and chat, and a single model underneath. Every conversation is a record in its own right, with participants, timing, the entities it is about, privacy settings and the outputs it produced.

Fitting into how people already work

Auron is built to follow the conversation rather than move it.

Meetings happen on the platforms you already use, and an agent joins as a participant to capture the discussion as it happens.

Calls happen through telephony. Where the audio is available, they are captured under the same model as meetings. An agent with a phone number assigned takes calls directly.

Chat is the complementary path inside Auron, for when a live session is not needed or when you want to work over the output of one.

Transcripts of conversations that happened somewhere else can also be uploaded and turned into a session, which is how a meeting with no agent in the room still ends up with signals attached to the right record.

Voice

Voice is the default capture medium because it carries the most context for the least effort. Spoken conversation contains the things that rarely survive into written notes: intent, trade-offs, implicit requirements, and the reasoning behind a decision.

Operating modes

An agent in a meeting runs in one of four modes, and whoever is in the meeting can change it as they go.

Listen is the mode for a conversation an agent should not join, such as a consultation or a negotiation. Recap earns its keep in the last minute of a call.

Voice as the evidence record

Voice sessions become time-coded transcripts with their session metadata. That record is what makes the outputs defensible: a summary can be checked against what was said, a signal carries the quote it came from, memory keeps its provenance, and an action can be traced back to the conversation that caused it.

Chat

Chat is not a different behaviour model, just a different way in. It is used for asking targeted questions against approved sources, refining a summary or a signal, drafting something structured, and deciding what should be kept as durable context.

Chats are captured as conversation records too, so they link to entities, respect access control, and can start follow-through.

Privacy and memory

Not every conversation should feed shared context. A standard conversation may update entity and organizational memory, subject to policy. A private conversation still produces its summaries, signals and drafts for immediate use, but is blocked from updating organization-wide memory. That lets a sensitive discussion benefit from an agent without widening what the organization knows.

Decision traces

Conversations are where the reasoning behind outcomes exists: what was decided, why, what constraints shaped it, what was promised. Everything else in Auron exists to extract and preserve those traces in a controlled way and attach them to the right entity. Over time that produces continuity: not just the current state of an account, but how it got there.

What a conversation pulls together

The agent brings persona, skills and policy. Knowledge and tools decide what context it can reach. Signals and summaries structure what happened. Entities and memory carry it forward, unless the conversation is private. Actions and automations turn it into follow-through.

Where this lives in the product

Sessions run in the console and on the phone, and every session is listed in Activities and in the phone's History tab.