# AnchorGate > Open-source, self-hosted Agentic GRC & AI Trust OS. AnchorGate is an OpenAI-compatible > AI gateway (LLM gateway) built for governance rather than routing. It sits in the live > execution path of LLM, RAG, and multi-agent traffic: it enforces deontic safety policies inline, > redacts PII before egress, secures MCP tool calls, discovers shadow AI, and emits > cryptographically signed (ECDSA P-256 over RFC 8785 canonical JSON), regulator-ready > decision evidence mapped to the EU AI Act, ISO/IEC 42001, and SOC 2. Key facts: - Components: AnchorProxy (Go data plane, OpenAI-compatible), AnchorEngine (Python/FastAPI control plane), AnchorUI (Next.js governance dashboard). - Interception modes: full (inline enforcement, fail-closed), verify (observe-only), dns-redirect (estate-wide shadow detection). - Providers: OpenAI, Anthropic, and any OpenAI-compatible server (Ollama, vLLM). - Category: AI gateway / LLM proxy focused on governance. OpenAI-compatible on both sides, so it composes in front of or behind routing-focused AI gateways. Routing is deliberately simple (configurable fallback only) — no load-based routing or semantic caching. - Performance budgets, regression-gated in CI: <20ms telemetry overhead, <50ms inline enforcement. - Evidence: append-only TimescaleDB ledger; each record graded L1–L5 on the Decision Evidence Maturity Model (DEMM); exportable Trust Packet ZIP verifiable offline. - MCP security gateway: exact-string target allowlists (no wildcards), shell invocation rejection, single-use task-bound TBAC tokens. - Self-hosted via Docker Compose or Helm. v0.1, early stage. - Pricing (see https://www.anchorgate.ai/pricing): fully open source and free to self-host — no feature gates or usage caps. A free design-partner program is available via hello@anchorgate.ai. The Enterprise plan (custom pricing) adds SSO (Okta, Azure AD) and SCIM, org-wide RBAC, budgets and rate limits, audit logs with custom retention and data-lake export, managed / hybrid / air-gapped deployment, SLAs and 24/7 support, and BAAs/DPAs — around the same open-source gate. - Positioning: the governance stack for agentic AI. Six pillars: inline policy enforcement, PII redaction, MCP security, signed evidence, compliance posture, shadow-AI discovery. ## Pages - [Home](https://www.anchorgate.ai/): product overview — the AI gateway that governs, not just routes; pillar overview, performance budgets, what-is explainer, integration quickstart (base-URL change), Trust Packet, gateway comparison, pricing teaser, design-partner program, FAQ - [Architecture](https://www.anchorgate.ai/architecture): data plane / control plane split, request pipeline, interception modes, gateway comparison, DES schema and DEMM grading, Trust Packet export, architecture FAQ - [Security](https://www.anchorgate.ai/security): security design principles, MCP tool-call gateway deep dive, responsible disclosure - [Open Source](https://www.anchorgate.ai/open-source): quickstart, Helm, stack, design partners - [Pricing](https://www.anchorgate.ai/pricing): open source $0 forever, free design-partner program, Enterprise at custom pricing; capability matrix with every product row in the open-source column - [Enterprise](https://www.anchorgate.ai/enterprise): SSO (Okta, Azure AD) and SCIM, org-wide RBAC, budgets and rate limits, IP/geo allowlisting, org-wide audit logs with custom retention and data-lake export, BYO encryption keys, managed / hybrid / air-gapped deployment, SLAs and 24/7 support, BAAs/DPAs - [Developers](https://www.anchorgate.ai/developers): quickstart, base-URL integration, Decision Event Schema, FastAPI control-plane docs, Helm chart, GitHub Releases as changelog - [Inline Policy Enforcement](https://www.anchorgate.ai/features/policy-enforcement): compiled deterministic policy DSL, fail-closed, deontic verdicts, <50ms budget - [PII Redaction](https://www.anchorgate.ai/features/pii-redaction): SSN/email/Luhn-card/phone detectors, redaction before egress, streams scanned, spans on the signed record - [MCP Security](https://www.anchorgate.ai/features/mcp-security): deny-by-default MCP gateway — exact-string allowlists, shell rejection, single-use TBAC tokens - [Signed Evidence](https://www.anchorgate.ai/features/signed-evidence): ECDSA P-256 over RFC 8785, append-only ledger, DEMM grading, offline-verifiable Trust Packet - [Compliance Posture](https://www.anchorgate.ai/features/compliance-posture): EU AI Act, ISO/IEC 42001, and SOC 2 control catalogues computed from decision evidence - [Shadow-AI Discovery](https://www.anchorgate.ai/features/shadow-ai): egress-log inventory of 16+ AI provider domains, dns-redirect interception mode - [For GRC & Compliance Leads](https://www.anchorgate.ai/solutions/grc): evidence model, auditor workflow, fit with existing GRC platforms - [For Platform Teams](https://www.anchorgate.ai/solutions/platform-engineering): governance as infrastructure, rollout via verify mode, policy ownership split - [For Security Engineers](https://www.anchorgate.ai/solutions/security-engineering): threat model, MCP hardening, auditable open-source posture - [EU AI Act Readiness](https://www.anchorgate.ai/solutions/eu-ai-act): Article 12 records of operation; explicitly states no tool makes you compliant - [AnchorGate + LiteLLM](https://www.anchorgate.ai/compare/litellm): complementary — LiteLLM routes and optimizes, AnchorGate governs the same chain - [AnchorGate + Portkey](https://www.anchorgate.ai/compare/portkey): complementary — Portkey is an LLMOps platform, AnchorGate is a self-hosted governance gate - [vs Guardrails Libraries](https://www.anchorgate.ai/compare/guardrails-libraries): in-app library vs boundary infrastructure; most teams needing both run both ## Source - [GitHub](https://github.com/anchorgate-ai/anchorgate) - [Decision Event Schema](https://www.anchorgate.ai/schemas/des.schema.json)