Skip to content

Agent Builder limits

Agent Builder respects organization plan limits and includes guardrails to prevent runaway automation loops.

Agent Builder checks your plan before confirming writes. If a change would exceed quota, the preview shows a blocking warning.

ResourceFreeStarterStandardProEnterprise
Assistants131030Custom
Knowledge bases11315Custom
Tools (API Request / MCP)API RequestAPI + MCPAPI + MCPCustom
Agent Builder sessions/day525100200Custom
Confirmed writes/session10255075Custom

Exact quotas match Plan limits and Plans compared.

⛔ Cannot attach tool: Starter plan does not include MCP tools.
Upgrade to Standard or use an API Request tool instead.

No write occurs until the conflict is resolved.

If Agent Builder detects a loop — repeated failed API calls, identical replans, or excessive tool tests — it pauses execution:

SignalThresholdAction
Failed API calls3 consecutivePause + show error detail
Replan attempts5 without confirmationStop and ask you to rephrase
Tool test calls10 in one sessionRequire explicit “continue testing”
Session duration30 minutes idleSession expires

Resume by starting a new request or typing “continue” when prompted after a soft pause.

Agent Builder sessions consume LLM tokens like assistant conversations. The dashboard usage panel includes Agent Builder usage under Configuration.

TipBenefit
Batch related changes in one confirmed actionFewer plan/review cycles
Use read-only queries for explorationNo write quota consumed
Avoid “test every SKU” loopsStays under circuit breaker

Agent Builder API calls share your organization’s global rate limit (1,000 req/min). Heavy automation should use the REST API directly with an API key.