General Limitations

Limit DescriptionLimit ValueNotes
Maximum size per input100Kb
  • In most cases, the input refers to the output of an ability execution. If this output exceeds the 100KB limit, the agent session will fail and may not respond as expected.

  • Recommendation: To avoid exceeding the size limit, workflow builders should transfer only the most relevant information from the ability execution to the agent input. Use workflow outputs to extract and pass concise, essential data
Maximum number of abilities per agent10An agent can execute the same ability multiple times within a single run
Maximum number of ability executions per agent run20
Maximum timeout for an Agent step14 days

Agent Depth

To prevent infinite loops, the system limits how many agents can call each other in a chain. This limit is known as the agent call depth. If more than 5 agents are invoked consecutively, either directly via workflow abilities or indirectly through subflows, the system halts execution and raises an error.
Note: Agent depth is tracked separately from workflow step count. The total execution chain (workflow steps + agent calls) can reach up to 15 steps combined.
Limit DescriptionLimit ValueNotes
Maximum allowed depth5 steps
Agent Depth Examples
Scenario DescriptionAgent Call ChainAllowed?
5 unique agentsAgent A → Agent B → Agent C → Agent D → Agent E✅ Yes
Same agent called 5 timesAgent A → Agent A → Agent A → Agent A → Agent A✅ Yes
6 agent invocations (loop pattern)Agent A → Agent B → Agent A → Agent B → Agent A → Agent B❌ No


Rate Limits

Limit DescriptionLimit ValueNotes
Maximum requests per minute (per tenant)25 requests/minute
Maximum agent data usage per minute (per tenant)0.5 MB
Maximum agent data usage per hour (per tenant)2.5 MB