A docket of what an agent did: thoughts, actions, questions, responses and errors, each timed. Questions waiting on you are highlighted.
A chronological stream for reasoning, searches, terminal work, and tool execution.
Showing 1–24 of 357
Card with a gradient cover, icon tiles, title, description and an optional author row.
Full agent chat pane: streamed markdown turns, tool chips, reasoning, and a composer with a model-controls slot.
Agents editing a document like multiplayer teammates: coloured live pointers with name, action and spend, suggestions typed in as tracked changes with accept, reject and Accept all, image resize, code edits, margin comments with a typing indicator, and a Table tab.
Many agents at a glance: a row (or card) per main agent with model, task, state, spend and clock, totals on top, and a sub-agent tree that unfolds under it with approve and retry right in it.
Live graph of agents and the tools they call: tree, horizontal and center layouts, light streaks per call, detail popover.
A relay of agents: who handed what to whom, with the context passed along, lost items, a light along the rail on each new hand-off, and forward reassignment by drag or keyboard.
One agent passing work to another — who asked, what was carried across, and what the receiving agent was not told.
The point where the agent hands control back: outcome, summary, next actions.
Where every agent is right now: a lane per agent, tasks growing into now, hand-offs and approvals on the block.
One agent at work as a card: current step with a clock and cost, a time strip, finished steps with tool logos, the exact command when it needs your OK, and a folded bar.
The steps an agent intends to take, checked off live: step states, sub-steps, elapsed times, a progress bar, animated plan changes, Approve plan / Edit plan and a one-line collapsed view. Reads the run store or plain props.
The checklist an agent writes for itself and then edits mid-run — steps added, skipped and re-ordered, without the list jumping under the reader.
Lets an agent ask for a decision: one to five single or multiple choice questions, a recommended option, Something else, number-key answers, and a record of the answers.
The run screen. It takes organisms as slots, so your app decides what goes where.
A presentation-only timeline for a multi-step agent / automation run: an ordered rail of steps with tool calls, outputs, errors, retry attempts, optional nested sub-stages, human approval checkpoints, and run-level lifecycle controls (cancel / resume). The app owns the run and every status; the component never executes an agent, generates reasoning, invents confidence, or claims a step succeeded on its own.
The activity log for a long agent run — collapsible success clusters, retry-linked events, approval gates in place, and a follow that respects the reader.
Status for an AI agent working on its own — running, waiting on you, completed or failed, with what it did last.
Shows the steps an AI agent is working through — done, running, waiting or failed — with a count in the header and rows that open to show what happened.
Tool calls as ruled rows, with state, tool, argument and duration. Expand a row to see its output.
A lead agent hands work to sub-agents that run side by side. Each gets a lane of tool calls, and you can scrub through the whole run.
Streamed answer with steps, sources, numbered citations that preview their source, copy, regenerate, feedback and follow-ups.
A file dropped into the chat uploads, then the assistant reads it page by page. Page citations in the answer light up the page they point at.