From LLM gateway to AI control plane.
One seam between your org and every AI capability. Shipping in quarterly drops — here's what's landing when.
Advanced Guardrails
Pluggable guardrails with pre-call, post-call, during-call, and logging-only modes.
-
Presidio PII detection -
Four execution modes -
YAML configuration -
Per-guardrail metrics
MCP Gateway (alpha)
Turn pLLM into the control plane for every MCP server in the org.
-
Server registry with manifests -
Per-tool allow-listing -
Health checks & circuit breakers -
Version pinning & blue/green
AD / Entra Governance
Native AD / Entra / Okta group sync. Policy-as-code driven by existing identity.
-
Group sync (AD, Entra, Okta) -
Model & MCP allow-lists -
Per-group budgets & quotas -
SIEM-ready audit streaming
Agent Registry
First-class agent artifacts with bound tools, guardrails, models, and AD scope.
-
Agent manifests (YAML) -
Bound MCP tools -
Canary & weighted rollout -
Per-agent cost attribution
Skills & Prompt Registry
Versioned skills and prompt templates with rollout controls and evals.
-
Semver prompt versioning -
Skill packs with intent triggers -
Built-in eval harness -
A/B tests across versions
MCP Gateway (GA)
Production-grade MCP control plane with sandboxed runtime and SLAs.
-
Sandboxed MCP runtime -
Multi-tenant isolation -
MCP-to-MCP composition -
Pre-built connector marketplace
Guardrails 2.0
Self-hostable classifiers, custom DSL, closed-loop eval from incidents.
-
Custom guardrail DSL -
Fine-tuned on-prem classifiers -
Incident → rule feedback loop -
Per-agent guardrail bindings
Cost Center & Observability
Unified OpenTelemetry traces, per-group cost rollups, chargeback exports.
-
OTel traces across the stack -
Per-group & per-agent cost -
Chargeback exports -
Anomaly detection on spend
Shape this roadmap.
Vote on features, file requests, or propose a new capability. The roadmap is driven by what our users actually need.