You can create Celigo AI agents that use AI models to process each record within your flow. For every record, the Celigo platform sends the mapped input data to the Celigo AI agent, which then provides a standardized response that can be mapped for subsequent steps in the flow.
Celigo AI agents are part of Celigo's agentic automation platform.
You can add Celigo AI agents to your flows as an import, and preview and validate the agent before using it in Production. Your Celigo AI agent can:
- Run tasks based on the instructions you provide.
- Call Tools or MCP servers to take actions in other systems.
- Return a consistent response envelope that downstream steps can map as:
- Text
- Structured JSON
- Blob
- Execute the following import components:
- Filters
- Mapper 2.0 / Transformation 2.0
- preMap and postMap hooks
- Standard error handling and run behavior
When your flow runs, the Celigo AI agent has the same order of operations as an import step. The platform reads the incoming records and processes them one at a time. For each record, any configured preMap hooks run first, followed by Mapper 2.0 mappings that prepare the input for the agent. The agent then runs using the selected model and settings. After the agent completes, any postMap hook is invoked, and the step produces a standard response that you can map to downstream steps. All activity is captured through the platform’s existing logs, metrics, and audit trails.
AI agent implementation lifecycle
Celigo supports the full implementation lifecycle of an AI agent, from defining what the agent does through testing, guardrails, monitoring, and troubleshooting in production. Each stage uses built-in platform capabilities, so you don't need external tooling to build, govern, or operate an agent.
- Define the agent's purpose and model. Decide what task the agent performs for each record, then select the AI model it runs on. See Create an AI agent for the agent settings, and "How and when to use AI models" in this article for guidance on choosing a model.
- Connect the agent to approved Tools or MCP servers. Give the agent access only to the actions it needs by attaching Tools or an MCP server that your team has approved. See Introduction to Celigo tools.
- Configure prompts, mappings, and supporting logic. Write the agent's instructions, use Mapper 2.0 to shape the input each record sends to the agent, and add filters or preMap and postMap hooks where you need custom logic.
- Apply guardrails and policy controls. Add guardrails to the flow to check records against safety, compliance, and governance policies before or after the agent runs. See Create a guardrail and Enhancing AI enablement with TRiSM: Trust, risk, and security in action.
- Test the agent before production use. Use preview to send sample records through the agent and validate the response before you run the flow in Production. Testing in a sandbox or non-production environment lets you confirm prompts, mappings, and tool calls without affecting live data.
- Review execution, tool-call, token, error, and guardrail logs. Monitoring is built in: every agent run is captured in execution logs that show each tool call, token usage, errors, and guardrail results. See Monitor AI agent and guardrail activity with execution logs.
- Troubleshoot failures and adjust the configuration. When a run fails or returns an unexpected response, use the execution logs and the flow's error management to find the cause, then update the prompt, mappings, model, or tools. Troubleshooting follows the same error handling and retry behavior as any other import step.
- Promote and operate the agent under defined governance controls. Move the agent to Production with the guardrails, model policies, and audit trails your organization requires. Model selection rules, guardrails, and TRiSM controls continue to apply at runtime, so the agent stays governed as models and policies change.
Navigate to your AI agents
Add the Celigo AI Agent import model type to run AI tasks on a per-record basis using standard import behavior and return a standardized response ready for downstream mapping.
-
To create your own Celigo AI agent as an import model type, navigate to AI studio → AI agent → Create AI agent.
If you have previously created an agent, select + Create AI agent at the upper right of the page.
-
Alternatively, you can add an AI agent while working in Flow builder. From the left navigation menu, select Build → Flows → select + and then select Add AI agent.
AI agents at a glance
Go to AI studio → AI agents to create or manage your agents and review activity.
The AI agents page lists every agent in your account so you can review status and key properties in one place.
- Name/Description – The agent's name and description.
- Last updated – Sort by newest or oldest.
- Actions – Select the Actions (...) menu for context-sensitive choices.
How and when to use AI models
Model selection controls which AI models your agent or guardrail can use and how those models run in a flow or API. It ensures that executions run under the correct account, respect model capabilities, and remain governable as models evolve or are retired.
You can select a model from the Celigo-provided list or from one of your own bring-your-own-key (BYOK) connections. The platform enforces all capability, policy, and lifecycle rules automatically so you don't have to manage them manually.
Why model selection matters
When you add an AI agent to a flow, the model determines output quality, performance, cost, and supported features. By controlling model selection centrally, you ensure:
- Only approved and supported models are used
- Inputs, outputs, and advanced settings stay compatible
- Runs draw from the correct quota and concurrency limits
- Retired or disabled models cannot be used accidentally
How model selection works
When you configure an AI agent or guardrail, you choose the model source first.
-
Celigo AI: You can select from a curated list of models managed by Celigo: These models run under a Celigo-provided enterprise-level account.
- OpenAI – gpt-5, gpt-5-pro, gpt-5-mini, gpt-5-nano, gpt-5.4, gpt-5.4-pro, gpt-4.1, gpt-4.1-mini (default), and gpt-4.1-nano
- Gemini – gemini-2.5-pro, gemini-2.5-flash-lite, gemini-2.5-flash, and gemini-2.5-flash-image
- Anthropic – claude-opus-4-8, claude-opus-4-7, claude-sonnet-5, claude-sonnet-4-6, and claude-haiku-4-5
-
BYOK: If you prefer to use your own connection (BYOK), you can select an OpenAI connection, Gemini connection, Mistral connection, xAI, Cohere, Hugging Face, DeepSeek, or Anthropic connection connector. Celigo automatically lists all models compatible and accessible on that connection.
Note
Supported models vary by provider:
- OpenAI – gpt-4.1 and later
- Gemini – 2.5 and later
- Anthropic – Sonnet, Haiku, and Opus
Service tier options (auto, default, priority) are available and control how requests consume capacity on your connection.
When to choose each model class
Choose a model class that balances quality, performance, and cost for your specific integration needs:
Pro: Offers the highest accuracy for complex, high-risk tasks where precision is critical, though it carries higher latency and costs.
Standard (Default): The recommended baseline for most integrations, offering a balanced mix of speed, cost, and output quality.
Mini: Optimized for speed and cost-efficiency. Ideal for routine or high-volume processes, but best avoided for heavy reasoning tasks.
Nano: The fastest and most affordable option. Designed for simple operations like data classification or extraction; not suitable for complex reasoning.
If a model becomes unavailable due to provider or policy changes, the platform prevents further saves that reference it and fails fast at runtime with a precise message and guidance.
Caution
If you attempt to access a feature the selected model does not support, the platform blocks the configuration or the run to protect your flows from unpredictable AI behavior in production. For some provider-specific settings, the provider enforces the limit instead, and you'll see the provider's error at runtime.