How to Use AI in Forest Admin, and How to Call Forest Admin with AI

Where is AI in Forest Admin?

As it’s Easter season, let’s say that, like Easter eggs in my garden, AI is in plenty of places in Forest Admin, surely more than you'd expect, sometimes buried in some bush feature, and sometimes in plain sight: AI is bolted in everywhere in Workflows, but with no flashy AI button.

This post explains how Forest Admin uses AI or lets your team use AI. This should help you make the most of it to run more effective, compliant workflows, whether that means delegating decisions to AI, plugging in AI agents, or connecting your own LLM instance to the platform.

How AI speeds up Workflows configuration in Forest Admin

Some types of task in the workflow builder use AI to handle the configuration work for you, before anyone runs the workflow. For instance, these two tasks:

Get Data.

Describe what you want in plain language ("Retrieve the KYC values from onboarding: name, date of birth, nationality, PEP status, sanctions result, risk rating") and AI structures the query.

Load Related Record.

Same principle. Write what you need ("Load the customer related to the alert"). AI maps that to the right collection, fields, and filters. No manual field or key mapping required.

In both cases, AI acts during setup, not execution. Once the workflow is saved, these steps run identically every time: no inference at runtime, no variability.

AI at build time vs. AI at runtime: two distinct modes

This distinction is worth understanding clearly, because it applies to every AI-powered step.

AI at build time is what Load Related Record and Get Data do. AI helps you configure the step. After that, it's out of the picture. The workflow runs deterministically: same logic, same fields, same filters, for every record. Reliable, auditable, reproducible.

AI at runtime is what the Decision step does. You write a prompt ("Based on the transaction history and screening results, would you recommend clearing this alert or escalating to a senior compliance officer?"), define the options ("All good, clear" / "Suspicious, escalate"), and set the Decision maker to AI.

When the workflow runs, AI reads the full record context and picks one of the options you defined. It cannot invent a new outcome. The choices are bound by what you configured. Every decision is logged with the reasoning behind it.

And of course, both modes can coexist in the same workflow: a Get Data step loads the context at build time; a Decision step later in the process uses that context to route the case at runtime. You control, per step, whether AI acts automatically or whether a human completes it manually. You decide where approval gates go.

AI-assisted, deterministic process execution: what it means in practice

If we had to name this approach, we'd speak about AI-assisted, deterministic process execution.

Each step is defined once and runs the same way every time. The full record context travels through the entire workflow. AI can handle a step, or a human can. Where the stakes require human sign-off, you add an approval gate. AI doesn't bypass it. Every AI decision is logged, with its reasoning.

For compliance operations, this is the difference between automation you can explain to an auditor and automation you have to apologize for. You define what AI touches, what it decides, and what it must justify.

Calling external tools and AI services from workflows via MCP connectors

Your workflows aren't limited to Forest Admin data.

Connector tasks let a workflow call external tools. This includes services with various level of AI integration: a sanctions screening API, a document verification provider, a fraud model, a custom LLM. If the service is MCP-compatible, it can become a step in your workflow. The Forest Admin connector library makes this a two-click configuration.

The result: a single, auditable workflow that pulls internal data, calls external AI services for enrichment or analysis, runs a decision step, and routes to a human reviewer where needed.

Letting AI agents access Forest Admin through the MCP Server

This works in the opposite direction too.

Forest Admin exposes your data, actions, and workflows to external AI agents through its own MCP Server. Any MCP-compatible agent can interact with your environment in two ways: reading data from your collections to use as context, and triggering actions or workflows directly.

In general, this means any AI agent platform like Dust or Claude can now process data and actions in Forest Admin.

For fintech teams, this means you can bring in AI agents for compliance operations and have them connect to your Forest Admin, making your datasources available and actionable to those agents. Agentic platforms like Bretton (AML and KYC investigations), Sardine AI (fraud and compliance), Arva AI (screening, AML, KYC/KYB), or Nebesta (complex support) can call Forest actions or workflows as part of their own processes.

Every call, to read data or execute a workflow or action, goes through the same permission layer as your UI and API. AI Agents act within the authorizations you've defined. And again, every action is logged and ready for analysis or audit.

Use our AI or bring your own LLM instance

All workflow AI-powered steps use the same LLM. And you can choose: using the LLM Forest Admin provides (currently free of charge), or connect your very own.

For many fintech and financial services teams, this isn't optional. They've gone through security review with one specific model. Their legal team signed off on it. Routing inference through a new provider isn't an option.

Bring Your Own LLM removes that blocker. As Alban from our engineering team put it: "Every AI call is redirected to the customer's agent. Zero requests on our side, no privacy violation."

This way, your data stays completely in your infrastructure. Your approved model handles inference. Forest Admin orchestrates the process without sitting in the data flow.

Get more details and see how to implement in our article: Bring Your Own LLM to Forest Admin.

One permission layer for every AI interaction in Forest Admin

Whether access comes from the UI, the API, or an external agent via MCP, it passes through the same permission system. Agents inherit the permissions of the account they act on behalf of: no more, no less. Every AI-initiated action is logged the same way a human-initiated one is.

AI augments your operations without creating new governance gaps.

Forest Admin is not an AI tool. It's the tool for AI.

You decide, per step, how much AI handles. External agents can read your data and run your workflows through the same controls your team uses. Every AI action is logged and explainable.

See! The eggs were there all along. Now you know where to look.

Ready to give it a try? Get in touch or book a demo.