Skip to main content

Agent Creation

Prerequisites​

  • Menu access: AI Studio > Workspace > Agent Creation.
  • Permissions: Read-Write access to create, edit, publish, list, delist, and delete Agents.
  • Data scope: Resource domain access in the current environment.
  • Model configuration: Connect LLMs under Model Management.
  • Referenced resources: MCP tools, knowledge bases, and Skills from Resources.

Overview​

The Agent Creation tab is where you create, debug, publish, and manage self-built Agents. An Agent is an AI execution unit for a specific scenario, combining reasoning, tool invocation, Skills, and knowledge.

Creation is a two-step flow:

  1. Basic info: Name, summary, description, build mode, icon, tags;
  2. Capability orchestration: Model, knowledge, tools, Skills, prompt—or a workflow canvas.

A debug preview panel on the right lets you test before publish. Published Agents appear under Resources > Agent > My and can be used in Sage AI and Automation.

image-20260821163847890

List quick classification and tags​

Workspace provides status quick filters (AND with the search box): All, Draft, Published, Pending review, Listed, Delisted, Rejected. Each shows a count (including 0). Default is All. Switching a filter shows only matching cards, sorted by update time descending.

Tags are required on create/edit so Resources can filter by tags after listing.

Value​

  • Two build modes: Autonomous Decision for complex reasoning; Workflow for fixed SOPs.
  • Visual orchestration: Workflow mode uses a drag-and-drop canvas.
  • Instant debugging: Right-side preview shortens the edit-test loop.
  • Version management: Each publish records version and change notes.
  • Shareable: Request listing for other domains to add from Public.

Build Modes​

ModeBest forCharacteristics
Autonomous DecisionComplex/unknown failures, multi-turn reasoningLLM-driven ReAct tool use
WorkflowFixed-step tasksStable visual node graph

Select the mode in Step 1; changing mode later may affect existing orchestration.

Use Cases​

Create and publish an Autonomous Decision Agent​

  • Audience: SRE, application ops, diagnosis engineers.
  • Approach:
    1. Create Agent → Step 1 (name, summary, tags, Autonomous Decision);
    2. Step 2: model, MCP tools, Skills, knowledge bases, system prompt (optional AI draft);
    3. Test in debug preview;
    4. Publish Agent with change notes.
  • Outcome: Status Published; available in Resources > My and Sage AI @ references.

Create a Workflow Agent​

  • Audience: Teams with standardized runbooks.
  • Approach: Step 1 → Workflow → arrange nodes (start/end, LLM, tools, knowledge, conditions) → test → publish.

Clone a built-in Agent​

  • Approach: Clone from Resources → edit in Workspace → publish as a new self-built Agent.

Request public listing​

  • Approach: Ensure Published with no unpublished drafts → Request listing → select domains → submit; Withdraw while pending; revise and republish if Rejected.

Operations​

Step 1 · Basic info​

  • Name * · Summary * (for intent matching) · Description · Build mode * · Icon · Tags *
  • Next → orchestration; Save draft or Cancel.

image-20260724161625000

Step 2 · Autonomous Decision orchestration​

  • Inference model, knowledge bases, system prompt, MCP tools, Skills
  • Debug preview (draft mode) on the right
  • Previous / Save draft / Publish Agent

image-20260724161700177

Step 2 · Workflow orchestration​

  • Canvas nodes: start, end, LLM, MCP tools, knowledge retrieval, conditions
  • Connect edges; test via preview or canvas run (per UI)
  • Publish Agent

image-20260724161723538

Save draft​

  • Saves progress as Draft; does not create a formal version.

Publish​

  1. Validation + content security check;
  2. Dialog: version (auto-increment, read-only) + change notes *;
  3. Status Published → syncs Resources > My; Automation tasks auto-upgrade.

Version management​

  • View history; only Publish creates versions.

Listing / delisting / withdraw / rejection​

ActionFrom stateNotes
Request listingPublishedRequires review rule; select listing domains
WithdrawPending reviewReverts to Published
Request delistingListedDelisting notes required
Handle rejectionRejectedRead banner → edit → publish → reapply

Listing domain rules (Agents):

  • Each listed domain must grant access to all referenced knowledge, tools, and Skills.
  • Domains failing any reference are not selectable.
  • Select all when every domain qualifies; after approval, new domains added later may also install this Agent (per UI hint).
  • If no domain qualifies, listing scope is empty—cannot submit.

image-20260724161818648

Delete​

  • Removes from Workspace and Resources; affects Automation and users who added a listed Agent.

Notes​

  • Summary and prompt quality affect Sage AI matching and review approval.
  • Listing uses the last published version if drafts exist—publish first.
  • Built-in Agents are not directly editable—clone instead.
  • Resource Marketplace details are read-only; edit in Workspace.
  • See Review Queue and Review Rules for approval flow.

Further Reading​