What we shipped, when.

MCP Server Inspection

Proxy and govern Model Context Protocol tool calls. Every MCP server call is inspected, policy-checked, and logged before execution.

Scoped Credential Vault

JIT tokens per tool, AES-256-GCM encrypted at rest, single-use. Credentials expire immediately after the tool call completes.

Computer-Use Recording

Forensic timeline of agent screenshots and actions. Full frame-by-frame audit trail for computer-use agents with tamper-evident hashing.

SIEM Export

Splunk, Datadog, Elastic, Sumo Logic, and generic webhook with HMAC signing. Push every event to your existing security stack.

Memory Guard

RAG and vector DB poisoning detection with ML scanning. Detects adversarial embeddings before they influence agent behavior.

Red Team Eval Suite

41-prompt adversarial corpus, regression tracking, CI/CD integration. Run evals on every deploy — know immediately if a model change breaks your defenses.

WorkOS SAML SSO + SCIM

Okta, Azure AD, Google Workspace, and OneLogin. Full SCIM provisioning so your IT team controls access from their IdP.

Invoice Billing

NET-30/60 payment terms, ACH/bank transfer, PO numbers. The procurement-friendly payment flow that enterprise finance requires.

Trust & Legal

DPA, MSA, and BAA templates ready to execute. Sub-processor list published at /trust. Designed for SOC 2 and GDPR compliance reviews.

Inline LLM Gateway

Streaming proxy for Anthropic and OpenAI with sub-50ms overhead. Sits in the critical path with fail-open semantics so your agents never go down because of us.

Trained Injection Detector

DeBERTa ML model plus 30 regex patterns, with async retroactive blocking. Catches prompt injection that simple pattern matching misses.

Output DLP

14 built-in patterns covering PII and secrets. Alert, redact, or block modes per pattern — tunable per agent and per environment.

Real-time Blocking

Policy engine runs before the upstream model call. Enforced at the gateway layer — not a webhook that fires after the damage is done.

Auto-instrumentation

One-line setup for LlamaIndex, AutoGen, CrewAI, OpenAI Agents SDK, Mastra, and Pydantic AI. Drop-in support for the frameworks your team already uses.