Platform

Overview

Ultron is an AI operating system for founder-led companies. A single chat surface drives seven named agents, a sandboxed shell, a background-jobs runtime, and a long-term memory layer. This page is the map to every surface you can touch as a user.

Updated today

What Ultron is

At a glance
Primary surface
The chat composer at the centre of the screen
Agents
Seven named: Cortex, Specter, Striker, Pulse, Sentinel, Amplify, Counsel
Skills
73 dispatchable units of work across the agents
Right panel tabs
Home, Activity, Builder, Computer, Brain, Deals, Profile
Tier picker
Minuet, Allegro, Forte (bottom-right of the composer)
Memory layer
Brain (Notes, Memories, Files, Agreements)
Long-running work
Background Jobs panel (donut indicator under the composer)
Connected tools
Gmail, Outlook, Calendar, Drive, Sheets, Docs, Notion, Slack, Reddit, more

Ultron is not a single model wearing a chat skin. It is a routing layer that picks the right agent, the right skill, and the right execution path for each user request. Most requests end up running a skill, calling a tool, generating a Canvas, or queueing a background job. A simple greeting bypasses all of that and just responds.

The user-facing surface is consistent. The chat composer accepts text, files, slash commands, and quick replies. The right panel changes shape based on what the agent is doing. Everything worth keeping is written to the Brain and indexed for retrieval next turn.

Surfaces

What you can see and touch.

SurfaceWhereWhat it does
Chat composerBottom-centre of the screenYour input. Accepts text, attachments, slash commands, and quick-reply chips suggested by the agent.
Tier pickerBottom-right corner of the composerSwitches between Minuet, Allegro, and Forte. Changes which models run and how the credit cost is calculated.
Workforce pickerSlash key in the composerType a forward-slash to open the full skill picker. Pick an agent (Cortex, Specter, etc) or a specific skill.
Background Jobs donutUnder the composerSpins while long-running work is in flight. Click to open the full Jobs panel.
Activity tabRight panelLive feed of what each agent is doing right now. Tool calls, sandbox commands, sub-agent fan-outs.
Builder tabRight panelCanvases the agent creates while working. Comparison tables, score cards, sequences, blueprints.
Computer tabRight panelThe sandboxed work environment. Shell access, file viewer, live preview of anything the agent is building.
Brain tabRight panelLong-term memory across Notes, Memories, Files, and Agreements. Searchable.
Deals tabRight panelThe pipeline view if you run deals through the agent.
Profile tabRight panelSettings, billing, brand voice, integrations.

How a request flows

Mental model for one full turn from your message to the result.

Every request goes through the same loop. Knowing the loop helps when you want a specific outcome. Skip ahead to the right surface page if you want the detail for that step.

StepWhat happensWhere it shows
1. SendYou type a message and submit. Attachments, slash commands, and quick replies all enter the same pipe.Composer
2. RouteThe platform reads the tier picker, decides which model class runs the turn, and picks the right tools.Invisible
3. PlanThe agent decides whether a skill is the right answer, whether to spawn a background job, or whether to respond inline.Activity
4. RunTools fire. Skills execute. Sandbox commands run. Sub-agents fan out. Each step writes an event you can watch live.Activity, Computer
5. SurfaceResults render in the composer reply, plus a Canvas in the Builder tab when the data has structure (a table, a comparison, a sequence).Chat reply, Builder
6. PersistAnything worth keeping is saved to Brain with the right tags so retrieval finds it next turn.Brain

Tiers

Three pickable levels, set at the bottom of the composer. Pick once, persists across the conversation.

TierBest forTrade-off
MinuetEveryday chat, quick lookups, simple drafts, content variations, idea exploration.Lower cost per turn. Fewer tokens in the output budget. Strong for chat, lighter on long-chain planning.
AllegroDeep research, multi-step skill runs, prospecting, complex outreach sequences, full canvases.Higher quality on structured output. Costs more per turn. Default tier for paid plans.
ForteStrategic reasoning, multi-document synthesis, debugging logic, architectural choices.Highest reasoning quality. Highest credit cost. Reserved for moments where the answer changes a decision.
Note
The tier you pick applies to the full turn. Slash commands, canvases, and skill runs all honour the picker. Detailed billing is on the Pricing page.

Where to go next

Pick a surface to drill into.

I want to know aboutRead this
The composer, slash commands, tools, narrationChat
The seven named agents and their skillsWorkforce
The donut, the jobs panel, long-running workBackground Jobs
Long-term memory: notes, memories, files, agreementsBrain
Canvases: comparison tables, score cards, blueprintsBuilder
The sandboxed Computer tabComputer
Tiers, credits, what each plan coversPricing