Security for the AI stack you already run.
Your AI agents connect to models, tools, APIs, and business systems. SAILLENT adds a runtime enforcement layer to covered execution paths, evaluating requests against explicit security policies before they proceed.
Keep your existing AI architecture. Add a policy enforcement boundary around the actions you need to control.
One enforcement layer. Multiple connection points.
SAILLENT is designed to sit between AI applications and the model or tool interfaces configured to pass through its runtime. It evaluates covered requests against policy and produces an allow-or-deny decision. The goal is to control what an AI system is permitted to do — not replace the models, frameworks, or business applications you already use.
Model requests and tool executions are separate paths. A model response does not automatically pass through the same enforcement point as a tool call — coverage depends entirely on which paths you route through the runtime.
Supported integration matrix
All listed integrations are Supported — tested against the current runtime release on documented paths. Coverage for any integration depends on which execution paths you route through the runtime.
| Integration | Status | Notes |
|---|---|---|
| Model Context Protocol (MCP) — covered tool interactions | Supported | Configured MCP tool calls are evaluated against policy before proceeding. |
 HTTP/JSON enforcement interface | Supported | Generic interface for compatible applications; per-application coverage depends on routing. |
| OpenAI-format model routes | Supported | Model API routing for inspection and enforcement on configured paths. |
| Anthropic-format model routes | Supported | Runtime policy evaluation on configured request paths. |
 Agent frameworks (LangChain, CrewAI, custom) | Supported | Supported when the framework routes model and tool requests through the enforcement path; coverage validated per deployment. |
Statuses reflect verification state as of October 2026. Request the current compatibility matrix — including tested runtime versions and known limitations — during an integration review.
How an integration works
Connect
Configure a supported application, provider, or tool interface to send covered requests through the SAILLENT runtime.
Inspect
The runtime processes the request information available to the configured enforcement path and evaluates the applicable policy.
Enforce
SAILLENT returns an allow-or-deny decision according to the configured policy and runtime behavior.
Record
The runtime records the policy decision and associated audit information according to the deployed configuration.
Designed for your existing architecture
- Your models remain your models. SAILLENT is an enforcement layer, not a replacement for an LLM provider.
- Your identity controls still matter. Continue managing user, workload, and service permissions through your existing access-control systems.
- Your policies define permitted actions. Enforcement depends on the rules you configure and the requests covered by the runtime.
- Your deployment determines data flow. A self-hosted runtime can process covered traffic in your environment, while model providers and separately hosted services process data according to their own configured roles.
Verify compatibility before deployment
Integration compatibility is specific to the runtime version, protocol, authentication model, request format, and deployment configuration. Before production deployment, establish:
- Which requests and operations pass through the runtime.
- Whether the required provider and protocol features are supported.
- How credentials and identity context are supplied.
- How denied requests and upstream errors are handled.
- What information is recorded in audit events.
- What happens during policy-source, network, and runtime failures.
We can review your architecture and identify the integration path to validate.
Frequently asked questions
Does SAILLENT replace our model provider?
No. SAILLENT provides a runtime enforcement layer around covered requests. Your model provider remains responsible for model inference.
Does connecting a provider protect every action an agent can take?
No. Protection depends on the execution paths and operations actually routed through the enforcement runtime. Direct or alternative paths outside that boundary require separate controls.
Does SAILLENT replace IAM or API gateways?
No. SAILLENT complements existing identity, authorization, network, and application security controls.
Can SAILLENT run in our own environment?
Deployment options depend on the product version and supported deployment configuration. Discuss your infrastructure and data-residency requirements with us before relying on a specific deployment guarantee.
How do we confirm an integration is supported?
Review the technical documentation and contact us with the provider, framework, protocol, runtime version, and operations you need to protect.
Bring your AI architecture.
Tell us which models, agents, tools, and systems you need to protect. We'll help identify the integration path and the coverage that needs to be validated.
Request an Integration Review
