What is IntentFrame?+
IntentFrame is a runtime authorization layer for AI agents. The agent proposes an action, IntentFrame checks that action against deterministic limits and plain-English policy, and returns a clear allow or block before the action runs. The agent proposes. IntentFrame decides. Only allowed actions run.
Is IntentFrame an agent framework?+
No. IntentFrame does not replace your agent framework. It sits around the action path. You can use it with your own agent, SDK-based agents, tool gateways, the Hermes Agent plugin, or an HTTPS authorization flow.
Does IntentFrame make my agent smarter?+
No. IntentFrame is a veto engine, not an optimization engine. Your agent developer makes the agent useful. IntentFrame makes sure a bad, confused, or compromised agent cannot cross the boundaries you set.
Do you stop prompt injection?+
No, and we say that directly. Prompt injection can still trick the model. IntentFrame's job is to stop the resulting action at the boundary. We cap damage at the boundary. We do not pretend to eliminate the cause.
Isn't this just prompting a second LLM?+
The naive version would be — two models reading the same prompt share the same blind spot. IntentFrame is different because the reviewer is in a different position than the agent. First, hard limits run with no AI at all and can block an action before any model sees it. When judgment is needed, the reviewer does not reread the agent's story: it gets an independent analysis of what the action would actually do, plus the context and policy the agent never had, and it treats the agent's stated reason as untrusted evidence. Allowing an action requires every layer to agree; any one layer can block. The agent is a requester; IntentFrame is a reviewer with an independent case file.
What happens when IntentFrame is uncertain?+
The safe default is to block or route to human review, not silently approve. Clear allow becomes ALLOW, clear violation becomes BLOCK, and anything unclear or risky is blocked or sent to a person depending on your workflow.
Which checks happen without any AI?+
Hard limits run deterministically before any model is involved: amount caps, allowed recipients and accounts, allowed action types, required fields, and obviously dangerous operations. A final deterministic check also runs just before an allowed action executes. AI is only used when a decision genuinely depends on meaning.
Where does my data go during a decision?+
Hard-limit checks run locally and send nothing out. When an action needs AI review, the action details and relevant context go to the model provider configured for your deployment. With private, self-hosted, or air-gapped deployments, that review stays inside your environment, so sensitive data never has to leave your boundary.
What is out of scope?+
IntentFrame does not make the base model correct, does not stop injection from changing the model's internal state, does not by itself catch abuse spread thinly across many individually-allowed actions, and does not govern actions taken outside the supported IntentFrame path. We would rather state the edges than oversell them.
Where can IntentFrame run?+
IntentFrame Cloud for a managed dedicated endpoint, private deployment inside customer cloud/VPC, self-hosted enterprise on Docker/Kubernetes, and on-prem or air-gapped deployments scoped during enterprise architecture review.
What if we are not using Python?+
Use the HTTPS authorization endpoint pattern. Your agent sends proposed tool calls to IntentFrame and receives an allow/block decision before executing the action. Python stacks can use the SDK directly.
Does IntentFrame make us compliant?+
No product makes an organization compliant by itself. IntentFrame provides controls and evidence that can support your compliance program: policy enforcement, credential separation, per-action decision records, audit trail, and deployment boundary options.
How is IntentFrame priced?+
Pricing depends on deployment model, the volume of governed decisions, integration scope, compliance requirements, and support level. You pay for the actions IntentFrame judges, not for seats.
Do you offer implementation help?+
Yes. Enterprise Implementation Services can include deployment, policy engineering, custom action wiring, compliance evidence mapping, and dedicated integration support. Our engineering team can deploy IntentFrame with you.
Why not just use monitoring?+
Monitoring tells you what happened. IntentFrame controls what is allowed to happen. If the agent already has authority, detection may arrive after the damage. IntentFrame sits before execution.
How do I reach you?+
Email
[email protected] or use the contact form at intentframe.com/contact to book a demo or architecture review.