Agent Studio (Builder, Templates & Runs)

Design reusable agent workflows at /builder, manage them at /my-agents, and open a saved template at /run/<id.

Date: 2026-09-30 Purpose, access and entry Design reusable agent workflows at /builder, manage them at /my-agents, and open a saved template at /run/ . Builder/My Agents are gated for paid subscribers and administrators. The community gallery at /community-templates is publicly browsable; browsing is not execution authorization. Inputs and workflow Choose an Official, Community or Personal template in Library, or use Build with AI with a description of your goal. Review the generated preview before Apply to canvas, or drag an agent archetype onto the canvas. In the Inspector set identity, prompt, permitted tools, model/control settings and declared inputs such as ticker. Connect agents for sequential/parallel execution; fan-in waits for upstream inputs. Use Test to validate configuration, then Save. Validation does not run live research. Select Run, fill required inputs and inspect the live trace and final answer. Approve each gated execution request only after inspecting its actual arguments. Approval tokens are bound to the action; a template cannot remove required human approval. Publish only after reviewing the safety check. Manage/clone saved templates through the library or My Agents. Platform agents in My Agents My Agents shows two groups. Your agents are the templates you deployed yourself (Launch / Edit / Undeploy / Previous runs). Platform agents are ready-to-run agents 7 Horns AI shares with every Agent Studio plan — currently Technical Analyst, Report Watchdog (grades the analyst report for a ticker and, after your approval, requests a regenerated report) and Discovery Agent (news theme radar when the ticker is blank, ticker impact when it is set). Each platform agent offers Launch, Clone to customize (makes a private copy and opens it in the Builder) and Previous runs; they cannot be edited or undeployed. Platform agents do not appear in the community gallery. Administrators can turn any of their own deployed agents into a platform agent from its card ("Share as platform agent"), choosing the public name and description; a copy is published and the admin's original stays private. Outputs and limits Outputs are templates, run traces and saved conversations where persistence succeeds. Tool availability follows the curated tool/source policy; selecting a tool or model is not a promise that every provider/model is enabled. Current execution may use the platform default model despite per-agent selection. Paper-trade tools operate simulations, not live broker accounts. Report regeneration and manually triggering a schedule are also execution actions in the template approval gate. Public templates cannot override server access checks. On narrow screens Library/Inspector use sheets rather than the desktop layout. Loading, errors and empty states An empty canvas/library needs a new or cloned template. AI generation and running have separate progress states. Invalid connections/inputs are configuration failures; unavailable tools, quotas or model failures happen at run time. A waiting approval is not a finished action. Do not assume an interrupted trace completed; inspect the saved run/thread before retrying.