AI governance & LLM ops
The same permission model, audit, and review workflow applies to AI agents as to human operators. Bring any model; the governance layer stays constant.

AI agents operating in regulated ops need the same controls every other actor has. Forest applies them structurally, not as an add-on.
Record-level audit for AI actions
Every agent action logs the model, the inputs, the reasoning, and the outcome against the affected record.
Scoped permissions for agents
Each agent connects with a role defining which data it can read, which actions it can trigger, and under what conditions.
Reasoning logs
Where the model exposes reasoning, it is captured with the action. Reviewers can see how the agent reached a decision.
Human review on high-risk actions
Actions can require human approval before execution. The approval chain is the same as for human-initiated actions.
Model-agnostic governance
Bring Claude, GPT, or open-source. The governance layer is the same regardless of model. Switching does not break the audit.
Regulator-ready exports
Filter by agent, model, customer, or time window. Export the full action and reasoning trail as an inspection bundle.


Built for AI in regulated ops
Permissions at the data layer
Each agent connects under a role defining what it can read and trigger. Enforced at the data layer; no prompt injection can expand it.
Audit at the data layer
Written at the moment the action happens, against the record it touches. Agents, humans, and automations share the same trail and format.
Approval chains at the data layer
High-risk actions pause for human approval before they run. The approval step is the same for agent-initiated and human-initiated work.
Vendor agnostic
Frequently asked questions
You still have question ?
Book a conversation with a forest expert
What does Forest log when an agent acts?
The model used, the agent's identity, the input data and prompt, the reasoning trace (where exposed by the model), the action taken, and the outcome. The log is attached to the record the agent acted on.
Can agents act without human review?
Within configured rules. For routine actions (matching configured patterns), an agent can execute on its own. For high-risk actions, a human approval step is required. Each action's risk and approval requirement is configured.
Can we switch model providers?
Yes. Forest's governance is model-agnostic. Switching providers does not change the audit shape or the permission model. The model name is recorded with each action so you can see which agent was acting on which model.
What about LLM reasoning that isn't exposed?
When the model does not expose its full reasoning, Forest records what is available (inputs, tool calls, citations, outputs). The audit reflects what the platform was able to observe.
How does this integrate with our GRC stack?
Forest forwards events to Drata, Vanta, Sprinto, Secureframe, Datadog, Splunk, or your warehouse via webhook or sync. The Forest log stays as the source; downstream tools get the events they need.
What does Forest not cover?
Forest is not an AI safety or red-teaming tool. It does not align models, evaluate model output for accuracy, or run safety tests. Those remain upstream concerns with the model provider or dedicated tools (Anthropic eval suites, OpenAI evals, Inspect AI).
Does governance work the same across model providers?
Yes. Forest's governance is structural. Whether the agent uses Claude, GPT, or an open-source model, the action recorded looks the same. Switching providers does not break the audit trail or the permission model.

