Skip to Content

Agent Studio Store & Catalog

The Agent Catalog (isagent.io/app/library) is the centralized repository for discovering, creating, and managing enterprise AI agents. It bridges business teams and developers, allowing organizations to package domain knowledge, system prompts, and MCP tool access into reusable, governed autonomous agents.

┌──────────────────────────────────────────────────────────────────┐
│  isA Agent Studio Store (Library)                                │
│  Search: [Filter by Category: Engineering, Ops, SCM...] [🔍]     │
├────────────────────┬────────────────────┬────────────────────────┤
│ 🤖 SCM Control Bot │ 💻 Fullstack SWE   │ 📊 Financial Auditor   │
│ Author: Operations │ Author: Dev Team   │ Author: FinOps         │
│ Tools: WMS, ERP    │ Tools: Git, Bash   │ Tools: SAP, Excel      │
│ [Add to Workspace] │ [Add to Workspace] │ [Add to Workspace]     │
└────────────────────┴────────────────────┴────────────────────────┘

Agent Architecture & Metadata

Every agent in the catalog is defined by a declarative specification containing four core layers:

  1. Identity & Persona:
    • name: Display name (e.g., “Postgres Performance Tuner”).
    • avatar: Visual monogram or brand icon.
    • description: Scope of expertise, limitations, and example prompts.
  2. Cognitive Configuration:
    • default_model: Default model provider and tier (e.g., claude-3-5-sonnet, gpt-4o).
    • system_prompt: Base instructions, operational constraints, and formatting directives.
    • temperature: Creativity/determinism threshold (0.0 to 1.0).
  3. Tool Entitlements (MCP Access):
    • Specific Model Context Protocol servers and tool functions the agent is allowed to invoke (e.g., mcp::github::*, mcp::postgres::query_readonly).
  4. Governance & Guardrails:
    • Rate limits, maximum token budget per run, and Cedar authorization rules.

Publishing & Visibility Tiers

VisibilityScopeAccess Controls
PrivateOnly accessible by the creator.Ideal for drafting, testing prompts, and debugging tool integrations.
Team / WorkspaceShared with specific teams (e.g., Engineering, SCM).Members of the workspace can invoke the agent in their sessions.
Organization-WidePublished across the entire enterprise directory.Centrally governed by organization admins with verified badge status.

Subscribing & Workspace Customization

Once an agent is added to your workspace:

  • Custom Knowledge Attachments: Attach project-specific documentation (PDFs, Swagger schemas, Markdown runbooks) to the agent’s persistent memory.
  • Prompt Injection Overrides: Tailor the agent’s tone or output language without modifying the upstream catalog template.
  • Team Audit Logging: Track token utilization, tool invocation counts, and user feedback ratings in the admin console.