Working with agents¶
An agent is an AI assistant that lives in your organization. It can hold a conversation, call tools, run skills, remember what matters, and work on tasks in the background — all within the permissions you give it.
Every organization starts with two agents:
- Janet — a general-purpose helper available to everyone on your team. She can answer questions, use your connected integrations, run your skills, hand work off to specialist agents, and remember context across conversations.
- Michael — an admin-only agent. Only owners and admins can see or talk to him. Michael manages the organization: inviting people, changing roles, creating and configuring agents, assigning models, and reviewing the audit log.
Starting a conversation¶
Open the app at app.kwirker.com, pick an agent from the left rail, and start typing. Press Enter to send.
The composer waits a beat
After you press Enter, the agent waits about two seconds before it starts working. Keep typing and your follow-up lines are bundled into the same request — handy when you think out loud across a few short messages. Any keystroke resets the timer.
Shut up and listen¶
Sometimes you want to give the agent a few paragraphs of context without it jumping in after every line. Toggle listen mode (the microphone-off control in the composer) and the agent quietly collects everything you send. When you turn it back off, your messages flush as a single prompt and the agent responds once, with the full picture.
You can also enter and leave listen mode just by saying so in plain language — the agent understands phrases like "hold on, let me explain" and "okay, go ahead."
Sessions¶
Each conversation is a session. Sessions are saved automatically and listed in the side rail, so you can step away and pick up exactly where you left off. See Conversations & sessions for the full guide.
Your work keeps running
Agents do their work on our servers, not in your browser. If you close the tab, lose your connection, or switch devices, the agent keeps going. When you reopen the session, you see everything that happened while you were away — nothing is dropped.
Files¶
Drag a file into the composer to share it with an agent. Every uploaded file is scanned for malware before the agent can touch it, and the same scan runs on anything an agent produces for you to download.
The built-in file viewer renders documents, images, code, and data right in the conversation, so you rarely need to download to take a look.
Voice¶
Speak to an agent with the composer's microphone, and have replies read back to you when an agent has a voice configured. See Voice for the available engines and how to choose one.
Delegating to specialists¶
Janet can route work to other agents on your team. Ask her to "have the coding agent open a pull request" or "get the analyst to summarize last quarter," and she starts a session on the right agent and hands off the task. Admin-only agents like Michael are never offered to non-admins.
You can create purpose-built agents from a few starting templates — a general-purpose helper, a coding agent, a data analyst, or a database assistant. Admins create these in Settings, or just ask Michael to make one for you.
Hidden agents¶
Your agent list stays focused by default. Some agents — for example, starter agents your organization doesn't use day to day — may be tucked out of view instead of cluttering the list. Hidden isn't deleted or disabled: you can still open a hidden agent directly, and it keeps working exactly as before.
To see everything, click Show hidden at the top of the agents page. Each agent that was tucked away shows a hidden badge. Click Hide hidden to go back to the focused view.
Memory¶
Agents remember useful facts across conversations — your preferences, ongoing projects, decisions you've made — and bring the relevant ones back when they matter. See How agents remember for the full picture, including how to ask an agent to forget something.
Permissions¶
What an agent can do for you is the intersection of your permissions and the agent's. If you can't do something, an agent can't do it on your behalf — and if the agent isn't set up for a capability, having permission yourself won't unlock it. Owners and admins manage these grants in Settings.
Guardrails¶
Every agent's traffic passes through a safety layer that scrubs sensitive data, screens prompts, and restricts outbound calls to an allowlist. See Guardrails for the tenant-wide defaults, per-agent overrides, and decision log.
Building your own¶
Ready to go beyond Janet and Michael? See Your first agent for a step-by-step walkthrough, or read Authoring skills to teach your agents new tricks.