Agents — Overview

Agents — Overview

Agents are reusable, prompt-driven components used across workflows for tasks such as captioning, QRG generation, and safety analysis.

• Profiles & overrides: Prompts and LLM knobs come from Core.AgentProfiles and are loaded at runtime via agentProfilesService.getCurrent(profileId).
• Caching: Agents commonly implement a small cache keyed by profileHash; when profiles change agents rebuild automatically.
• Where agents live: See src/mastra/agents/ for domain-specific agent builders (e.g., k12, tap, reasoning-engine).

Index (by surface)

• **K12** — K12 Agents page contains K12-specific agent patterns for EOP editing and QRG generation.
• **TAP** — TAP Agents page contains TAP-specific captioning, content safety, and RAG query agents.
• **Reasoning Engine** — Reasoning Engine Agents contains multi-agent analysis pipeline agents.

Want per-agent prompts, registry IDs, and example calls? The detailed agent reference contains registry ids, profile ids, and common usage patterns.

Operator & admin tooling

Platform operators can manage agent profiles and overrides via the Admin surface. The Admin tools page documents runnable endpoints such as admin-update-agent-profile, admin-list-agent-profiles, and admin-agent-profile-history and includes examples for seeding prompts and auditing profile history prior to agent startup. For operational guidance about where prompts are stored and how agents read overrides, see this page and the Admin tools examples.