Platform · Console
Ask for the work. Approve what matters.
Console is where your team works with its agents. Say what you need in plain language. Kate, the orchestrator, answers what she can and hands the rest to the right specialist. Steps that need a person wait for one, and every decision comes back to the conversation that raised it.
01The conversation
A conversation that ends in finished work
Ask a question and you get an answer. Where command plans are switched on, ask for work and you get a plan: the goal, the constraints, the inputs, the agents that will run each step and the files each one will open. Nothing runs until you confirm it.
Once it runs, the conversation shows what is live and what is waiting on you. When a step needs a decision, the approval arrives in the thread with the exact thing being approved, and the outcome comes back on its own. You never have to go looking.
02Who you are talking to
One front door, the whole team of agents behind it
You talk to Kate. She knows what every agent can reach, assigns the work, and relays the
questions and results. Type @ to address a specialist directly, or
/ to start one of your governed skills.
Orchestrator
Kate
Answers what she can, routes the rest, and relays questions and results so you stay in one conversation.
Specialist
Analytics agent
Queries, dashboards and reports over your connected data, with every figure checked against a real result.
Specialist
AI forward-deployed engineer
Turns a request for a new workflow into a build brief you approve before anything is built.
Specialist
Ontology Author
Drafts the change to your operating model, delivered to Studio as a proposal for review.
Specialist
Code Author
Handles code work for your organization. The code tools it builds run under the same approvals as any other action.
Your fleet
Your own agents
The workers and supervisors deployed for your workflows, reachable by name or through a plan.
When Kate hands work to a specialist, its steps stream into your conversation as they happen. An agent that cannot reach something tells you what it is not granted and where to go instead.
03From request to result
The plan is the gate
Command plans are switched on for each environment. Where they are on, every piece of work Console starts takes the same road, and a person holds each gate on it.
01
Ask.
Plain language, with files attached. Kate answers or routes, and asks one clear question when the request is ambiguous.
02
Review the plan.
Goal, constraints, inputs and the agent on each step, all editable. A step missing an input is refused with the input named.
03
Watch it run.
Live status per run. A task waiting on a person is never shown as merrily executing.
04
Decide in the thread.
Approve one or several at once. Approved, denied or expired, the outcome comes back to the conversation.
04The conversation ledger
Everything the conversation caused, on the record
A conversation keeps one record of what it set in motion, and the agents answer from that record rather than from memory. It opens in a side panel beside the thread, so the evidence behind a decision sits next to it instead of in a window on top of it.
Hand an agent a spreadsheet and it can turn the rows into records without the contents passing through the model, and each record keeps a link to the file it came from. Documents come back as real files, not text pasted into the chat.
- Decisions Approvals and proposals, with the evidence behind each one
- Plans Every plan proposed, confirmed or superseded
- Runs and tasks What started, what is live, what is waiting on you
- Documents Files handed over, and real files handed back
- Reports Dashboards and reports the conversation produced
- Pull requests Changes to code and to your operating model
05Governed by default
An agent never reaches further than you could
Asking an agent is not a way around an access control. The platform works out who you are from the turn itself, never from anything the agent says, and checks every action against your own permissions before the action's own gate is considered.
A live turn runs as you.
Reads are limited to what you may see, and writes through your connected integrations stop for an approval that shows the exact payload before it goes out.
Dispatched work runs inside its limits.
Work that carries on after your turn ends is bounded by the agent's own scope, trust level and approval gates, and reaches external systems only through shared connections.
Every claim is checkable.
A figure in an Analytics reply that no result supports is flagged under it, and a reference to a plan or run that is not on record is marked as such.
06FAQ
Common questions about Console
Is Console just a chat window?
No. The conversation is how you ask, but what comes back is governed work: live status while it runs, approvals that come to you with the exact thing being approved, and a ledger of everything the conversation caused. Where command plans are switched on, you also confirm a plan before anything runs.
Can an agent act without my approval?
Any step your operating model marks for approval stops and waits for a person, with no timer that fails the run. A live turn runs with your own access, so an agent cannot reach data or actions you could not reach yourself.
Which agents can I talk to?
Kate, the orchestrator, by default. She routes to specialists such as the Analytics agent, the AI forward-deployed engineer, the Ontology Author and the Code Author, and to the agents deployed for your own workflows. Type @ to address one directly, or / to start one of your governed skills.
Who in my organization can use it?
Everyone from Viewer up can open Console. Operating tasks and deciding approvals need Editor or above, and every action passes the same permission check as the rest of the platform.
Does it work on a phone?
Yes. On a phone Console runs full screen, your work list opens as its own panel, and approvals, plans and documents take the whole window when you open them.
Bring a workflow. Watch a conversation run it.
See Console on your own documents in a live demo: the request, the plan, the run, and the one decision that should come to a person.