Platform specialists
Tealfabric ships two platform specialists that Trace AI can delegate to without a tenant administrator creating Autonomic Agents by hand. They appear in the Trace subagent catalog as soon as Autonomic Agents and Trace subagents are enabled.
They are not chat agents and do not appear in the in-app agent picker. They are also not Library model Autonomic Agents (sales watchers and similar business templates that you install as drafts).
Document information
| Field | Value |
|---|---|
| Canonical URL | /docs/07_ai-agents-and-mcp/27_platform_agents |
| Version (published date) | 2026-09-13 |
| Tags | ai, trace-ai, subagents, autonomic-agents |
What you get
| Name | Catalog slug | What it is for | What it must not do |
|---|---|---|---|
| Platform bulk reader | platform-bulk-reader | Answer one question about many tenant files or context resources. Returns structured bullets. | Create, update, publish, or execute anything. |
| Platform code specialist | platform-code-specialist | Generate ProcessFlow step code and/or WebApp content from a spec and a required reference, validate with read-only tools, return artifacts to Trace AI. | Create, update, publish, or execute platform resources; invent process_id / webapp_id. |
Trace AI creates unpublished ProcessFlow and WebApp records after validation succeeds. Publishing and execution stay Trace's separate steps.
Trace AI stays responsible for planning, asking you questions, confirming high-impact mutations, and the final answer. The specialist runs in the background on its own hidden session.
How to use them
Ask Trace AI in ordinary language, for example:
Summarize these three documents without loading them all into this chat.
Write ProcessFlow step TypeScript for this spec, matching the existing Order step, and validate it. Do not publish.
Trace loads skill_id: subagents, reads list_subagent_catalog, and should pick a platform slug when it fits. Job hints (SUGGESTED THIS TURN) for intake forms, WebApp login, and ProcessFlow-plus-integration name platform-code-specialist.
You do not need to create these agents under Agents first. The first time Trace delegates to a platform slug, Tealfabric compiles a tenant Autonomic Agent from the official template.
Requirements
The deployment must have:
AUTONOMIC_AGENTS_ENABLED=1TRACE_SUBAGENTS_ENABLED=1
Trace AI subagent tools still require a tenant-administrator role.
After the first run
Once a specialist has been used, it shows up under Agents like any other Autonomic Agent (same slug). You can then:
- inspect execution logs for that delegated run;
- set Model on that definition if you want a catalog model other than the tenant autonomic default;
- change concurrency or disable the agent.
If you create your own Autonomic Agent with the same slug (platform-bulk-reader or platform-code-specialist), your row replaces the official catalog entry for that tenant. Do not recreate those slugs unless you intend to replace the platform template.
Which model they use
Both templates ship with no pinned model. A run then uses the tenant autonomic LLM context (the same default as other Autonomic Agents — currently gpt-6-luna unless your tenant overrode that context).
To pin a catalog model for one specialist, edit that Autonomic Agent after the first delegation. Changing the official template in the product repo only affects new provisions, not rows that already exist.
How they differ from other agent types
| Kind | How it appears | Who creates it |
|---|---|---|
| Platform specialist | Trace subagent catalog (source: platform) | Product repo; first Trace delegation materializes the tenant row |
| Tenant specialist | Catalog only after you enable Allow Trace AI to delegate | You, under Agents |
| Library model agent | Not in the catalog until you install and activate it | You, via Trace model-agent-import (draft) |