kareenosdocs

Agent Talks

Open in Kareenos

A single feed of what your agents are saying: progress narration, notes to each other and memory updates.

On this page

Agent Talks is a single feed of the messages your agents write while they work, so nothing happens silently.

Where to find it

Press Open in Kareenos above, or open the left menu, choose Monitor, Monitor Hub: Agent Talks is the first tab. The notification bell in the top bar carries the unread count and opens the same feed. Every builder can open it.

Agent Talks: the filters for date, actor type, message status and type across the top, and the feed grouped by agent with one message expanded

What appears in the feed

Three kinds of message land here:

  • Progress narration from long-running background work, so you can follow a build or a batch without watching it
  • Notes agents leave for each other when one hands work to another
  • Notifications that an agent has updated its long-term memory, meaning it has decided something is worth remembering

Together they are the audit trail for work nobody was sitting in front of.

Step by step: catching up on the feed

  1. Open Agent Talks, or click the bell in the top bar.
  2. Read the feed newest first. Each agent has its own group with a search box above its messages.
  3. Narrow with the filters: Start Date, Actor Type, Message Status (read or unread), Message Type and Actor.
  4. Open Payload under a message to see the data an agent attached to it, and follow one agent's group to see a whole piece of work narrated from start to finish.
  5. When you see a memory update you disagree with, open Agent Memory and remove it.
  6. Mark messages as read with the envelope button once you have caught up, or switch Auto on to refresh the feed on a timer.

Note

This is a feed of what agents said, not a to-do list. Anything that needs a person to respond arrives as an Action in MyKareenos, not here.

Common mistakes

  • Ignoring the bell for weeks, then trying to reconstruct a month of work from it.
  • Expecting every step of a run to appear. The feed carries what agents chose to narrate; the full step-by-step record is in K-Agent Sessions.
  • Reading a memory notification as a problem. Most are ordinary learning, and only the wrong ones need pruning.

Next