Deep technical reference for engineers integrating with IntentGate. API surface, deployment architecture, and integration recipes for common SIEM and observability stacks.
Executive overview: the authorization pipeline, what ships, architecture, compliance mapping (AI Act Annex IV, DORA, GDPR, NIST AI RMF, SOC 2), OWASP for LLM and Agentic AI, pricing, and pilot terms. Built to forward to a CFO, CISO, or Risk Committee.
The full deployment set: install and getting-started guides, the Operator Manual, Operations Runbook, Operating Model, and the operating processes (actor swimlanes and RACI). Dedicated customer work, provided behind a login when you start a pilot.
Every endpoint the gateway exposes. Runtime path (/v1/mcp,
/v1/tool-call) plus admin path
(capabilities, policies, audit, approvals, tenants).
Authentication, error codes, request and response shapes.
What runs where, what talks to what. Components, request lifecycle, audit chain, multi-tenant model, three deployment topologies, trust boundaries.
The controls the MCP firewall enforces at runtime, schema sanitizer, intent enforcement, runtime limits and circuit breakers, kill switch, credential brokering, deception, session rewind, and proof-of-intent. Each with a diagram.
Recipes for common enterprise topologies. Each page is vendor-agnostic and links to deployable artifacts in the gateway repo.
Audit events into Microsoft Sentinel (hot lane, 12 months) and Amazon S3 + Athena (cold lane, multi-year). Native sinks, Glue table DDL, KQL query pack.
Self-hosted authorization service. Go binary, embedded OPA, multi-tenant, tamper-evident audit chain.
Apache 2.0. CI, signed releases, SBOM + provenance.
ghcr.io/intentgate-app/intentgate-gateway:1.7.1 Multi-arch container image. Distroless base.
Deploys gateway + extractor. Postgres-backed audit.
Standard Helm. About three minutes.
One signs tokens, one gates the admin API.
Gateway, extractor, Postgres-backed audit. No CRDs.
One health endpoint confirms the pipeline is live.
One credential per agent. Issued by the gateway.
Subject, tenant, and the tools it's allowed to call.
TTL, max calls, optional cost ceiling.
Shown once. Stored like any secret. Revocable any time.
Want the commands? GitHub for source.
Four end-to-end runnable guides in the open-source gateway repo. Each is a single milestone you'll hit when bringing IntentGate into your environment. Copy-paste from a fresh laptop.
One docker run in-memory, mint a token, watch the
capability check fire on a tool the token doesn't grant. The
shortest path from zero to seeing a real authorization decision.
Write Rego, push it via /v1/admin/policies/drafts +
promote, test allow / escalate / block paths against the live
engine. Includes the dry-run pattern and common rule shapes
(time-of-day, per-agent overrides, step-up required).
pip install intentgate, three-line
Gateway() construction, typed exception per check.
The LLM-loop-with-error-feedback pattern and
attenuate() for sub-agent delegation.
GET /v1/admin/audit filter parameters, SOC playbook
queries (blocks in last hour, all activity by agent), chain
verify endpoint, the daily-evidence-pack pattern for SOC 2 CC7.2
and EU AI Act Article 12.
The four guides cover the developer and SOC paths end-to-end: zero to a working policy, a wired-up agent, and queryable audit. For production deployment patterns, Helm-based HA, monitoring + alerting, day-2 operations, incident playbooks, full environment variable reference, the 23-page deployment runbook is the canonical document.