No request bodies in analytics
Source content and task text are processed for optimization without being written to usage analytics. Telemetry records token counts, latency and status—not source text. TrimLayer does not receive a model completion.
Enterprise Security & Governance
Your model access & data
stay under your absolute control.
TrimLayer's optimize-only workflow accepts source text and a project key, returns optimized context, and uses encrypted short-lived recovery only when enabled.
ISOLATED CREDENTIAL BOUNDARY
TrimLayer Project Keycl_live_••••••••••••
Only a TrimLayer project key is required. Provider credentials are not collected or accepted by the optimize-only API.
Core Guarantees
Minimized by default.
Recoverable by choice.
Normal optimization is transient, usage analytics remain content-free, and exact recovery uses encrypted short-lived storage only when a customer enables it.
Your provider key stays out
TrimLayer never needs, collects or accepts an OpenAI, Anthropic, Gemini or other model-provider credential in the optimize-only workflow. Your application makes the next model call directly.
AES-256-GCM encrypted store
When exact recovery is enabled, omitted original context is encrypted with tenant-scoped keys in a temporary store and deleted when its short retention period expires.
Fail-closed safety model
Any uncertainty, schema mismatch or time budget breach returns the original source context for your application to use unchanged.
Threat Model & Controls
Defense-in-depth across
the context lifecycle.
Architectural controls separate project authentication, tenant data, optimization decisions and the downstream model call that remains in your application.
In-Flight EncryptionTLS protects requests in transit; project secrets are never returned after creation
PostgreSQL RLSRow-level tenant isolation with strict least-privilege runtime credentials
No Model-Provider EgressThe optimize-only API returns context and does not route a request to a model provider
Drift Kill-SwitchReal-time distribution drift detection disengages optimization automatically
Data Governance Matrix
How TrimLayer compares
to conventional proxies.
A transparent breakdown of credentials, content processing, optional recovery and downstream model-call boundaries.
CONTROL AREATRIMLAYER BEHAVIORSECURITY GUARANTEE
Provider API KeysNot collected or accepted by the optimize-only APINever needed
Source & Task TextTransient optimization with no request body in usage analyticsNot used for training
Recoverable Original ContextAES-256-GCM encrypted temporary store with strict TTLTenant-isolated
Downstream Model CallsPerformed by the customer application with its existing provider credentialsOutside TrimLayer
Transformation UncertaintyAutomatic fallback to the original source contextFail-closed safe
Telemetry & Usage MetricsToken count deltas, latency, and status codes onlyContent-free
Compliance & Operations
Enterprise-grade compliance
and deployment options.
Engineered to satisfy stringent information security audits, privacy requirements, and enterprise governance.
Designed for security reviews
The architecture is designed to support SOC 2 and HIPAA-aligned deployment requirements. Customers remain responsible for their own compliance assessment and configuration.
Clear review boundariesIsolated Tenant Boundaries
Every organization and project operates in an isolated tenant scope governed by PostgreSQL Row-Level Security and cryptographic key isolation.
Complete tenant separationDedicated VPC & Self-Hosting
For regulated workloads requiring on-premise or isolated cloud boundaries, TrimLayer can be deployed inside your own AWS, GCP, or Azure VPC.
Private cloud deploymentLast updated: September 18, 2026
Get Started With Confidence
Evaluate before using optimized context in production.
Use representative source-and-task requests in the playground, compare the returned context and decide in your own application when it is ready for a downstream model call.
Open the consoleTechnical Documentation
Read our formal Correctness Contract.
Dive into the mathematical bounds, extractive proof contracts, calendar and typed validation rules that safeguard your AI workloads.
Explore Security Docs