Cortex

Olano's autonomous cognition engine. When Cortex is on, the agent does useful background work between conversations: distilling what it learned into memory, tidying its own notes, expanding ideas, and generating proactive suggestions. This tab shows what it has been doing and what is scheduled. It is off until you turn it on: the Cortex engine switch is inside Cortex settings below, for the whole deployment or for one agent (see Cortex and Pulse).

Everything you can change about it sits in one card, Cortex settings, at the top of the tab. It is folded away by default — the line beside its title is the live state (on or off, the pace, and today's deep-run count), so you only open it when you want to change something. Inside are three parts: How it runs, the switches; Phases & schedule, what runs when; and Creative direction, what the agent should aim at when it thinks on its own. Fleet-wide defaults live in Config; per-agent overrides live here and in the agent's own settings (see Cortex and Pulse).

The Brainstorming, Ideas, Prototypes and Projects tabs are the four stages of one funnel (see the funnel). A card that the Rate & Promote phase has reviewed carries its score as a badge (7/10; hover for the recommendation and the reason), and the drawer shows the reason and the suggested next step under the status line. Create project on a card jumps the item straight to a project; the agent's own cortex_promote moves it one stage at a time. A promoted card shows a link to the folder it became. The Phases & schedule list has a row for Rate & Promote like every other phase, with Run now to rate on demand.

Everything Cortex does in the background keeps running when you look away. Create project, Think now, Optimize workspace, Optimize skills and a phase's Run now each start a run on the server and open a panel that follows it: what it is doing right now, how long it has been going, and every step it has taken (each tool it called, what came back, what the model said). Closing that panel, leaving the page, locking your phone or reloading does not stop the run. While a project is being created its card reads Creating project… with the step it is on, and its button opens the same run again rather than starting a second one; a phase that is running shows Running in its row for the same reason. Pressing a button for something already running (from another tab or another device) always joins that run. When it ends you get a short notice, the card moves on (a promoted idea goes to Promoted, the new project to Projects), and a run that failed within the last hour leaves a Last attempt failed line on its card with a View log link. Stop in the panel ends a run early.

Runs, next to Refresh at the top, lists every run for every agent you can see: what is running now (the button shows a count while anything is) and the history before it, including the phases Cortex runs on its own schedule and the tasks you approved in Pulse. All, Running and Failed narrow the list, the agents filter beside the tabs narrows it too, and opening a row shows that run's full step log. A run that was going when the service restarted shows as Interrupted. The most recent 200 runs per agent are kept.

Two things on this page have a different scope, on purpose. The buttons along the top — Think now, Optimize workspace, Optimize skills, Activity — and the settings card act on one agent: the one named under the title, which you change with the agent dropdown in the page's top bar (the same dropdown Notes and Memory carry). The tabs below — Pulse, Brainstorming, Ideas, Prototypes, Projects, Learned Skills, Self-Improve, Self-Checks — show every agent's items at once, so reviewing what the fleet has come up with no longer means switching the dropdown once per agent. Every card carries a small label with the agent's face and name it belongs to, and the buttons on that card act on that agent (a Chat switches you to it). The agents filter beside the tabs — a dropdown with a checkbox per agent, a search box, Select all and Clear — narrows the tabs to the agents you tick; the line next to it says what is showing ("Showing items from all agents", "from 2 of 8 agents"). Every agent is shown by default, an agent created later is shown too, and the choice is remembered on this device. The counts on the tabs follow the filter.