Note
Celigo AI agent is available to all Celigo platform users for free until the end of June 2026.
AI agents actions are counted when they are executed as part of flow runs, and AI credits measure how much processing/compute are used by the agent actions. In the Celigo platform, usage is tracked on the Subscription page. It shows:
-
AI agent actions – Measures the agent actions Each tool execution is counted as a single action
-
AI credits remaining – Measures the AI processing utilized
AI credits are only consumed when an agent performs AI tasks such as model reasoning, tool calls, or response generation. Actions like creating, editing an agent, or saving configurations do not use credits, but they will be used when you run a test that triggers AI.
The Celigo platform will not prevent you from invoking your agents if you exceed your AI actions entitlements. You will receive email notifications when you are nearing, at, and over your limit.
Note
Monitor your usage of AI credits and AI agent actions. If your balance is low, you can request additional entitlements.
The Celigo platform will prevent you from invoking your agents if your AI credits balance reaches zero or lower. You will receive email notifications and error messages when you are nearing, at, and over your limit.
Warning
An error message insufficient AI agent prepaid credits is displayed when your AI credits balance reaches zero.
AI actions are consumed during agent execution when an AI agent performs any tool calls. Tool calls include:
-
LLM native tools – Tools like image generation, web search, or others provided by the LLM.
-
MCP tools/resources – Calls to tools via an MCP Server or retrieval of resources via an MCP Server.
-
Celigo tools – Tools added directly to an agent.
-
Retries and loops – Repeated failures and re-attempts during tool calls are also counted as an AI Action.
Note
AI actions consumed during testing/review or incurred in non-production environments are not counted.
AI credits are used when AI agents are executed in flow runs.
Credit usage typically increases when run include:
-
More AI steps – Multiple reasoning or generating calls in a single run
-
Larger inputs – Long prompts, big context, large payloads/logs/documents
-
Larger outputs – Wordy responses
-
Retries and loops – Repeated failures and re-attempts
Even for the same agent, credit usage can differ because two runs consume different credits. This happens if the runtime behavior vary due to:
-
The input data differs (size/complexity)
-
The agent takes different paths (more steps, different tools)
-
Failures trigger retries
-
Prompts result in longer or shorter outputs
In general, when more data is processed and more reasoning steps are involved, the credit usage increases.
Use these practices to keep usage manageable:
-
Set budgets or limits where possible. Limit usage per run (for example, max tokens) or over a time period to avoid spikes.
-
Keep prompts and context concise - Include only what is necessary. Avoid passing large logs or datasets unless required.
-
Use structured inputs instead of freeform text to reduce ambiguity.
-
Restrict output length by requesting concise formats (such as bullets, JSON, or short summaries) unless detailed responses are required.
If runs consume more AI credits than expected, the most common causes might be:
-
Multiple model calls within the run (multi-step loops, repeated planning)
-
Large context including extensive payloads, logs, documents
-
Prompt that requires detailed or lengthy response
-
Retries triggered by tool failures or validation errors
Q. Do AI credits get consumed if the agent doesn’t use AI during a run?
A. No. Credits are consumed when AI work is actually performed during execution.
Q. Are AI credits the same as agent actions?
A. No. Actions represent tool executions (what happened). Credits represent AI compute usage (how costly it was.
Q. Can I predict the exact credits before running?
A. Not exactly. You can estimate by testing an agent and reviewing the credits consumed. This provides a rough baseline especially if your input records are of similar size and complexity.