Back to Home

Roadmap

What We're Building

The Hub is actively growing. Here's what's live now, what's coming next, and our long-term vision.

Phase 1Live

Text & Vision Firewall + Budget Enforcement

  • 30+ PII entities with configurable actions (block / redact)
  • OCR-based image PII scanning (Base64, max 5MB)
  • Context-aware full conversation history DLP
  • Smart routing + managed credits + BYOK hybrid billing
  • Policy versioning with restore & audit trail
  • Custom regex IP Guard rules
  • Prompt injection / jailbreak heuristic blocking
  • Multi-provider failover with automatic fallbacks
  • Image generation support (FLUX, DALL-E, Stable Diffusion)
  • Streaming chat completions (SSE) with full DLP on input
  • Per-request token caps with pre-flight balance enforcement
  • Real-time per-project cost dashboard with per-model breakdown
  • Wallet balance enforcement — requests blocked when balance hits zero
Phase 1.1Shipped

Developer Experience & Compliance

Integrate Faster, Stay Compliant

A native SDK, real-time security alerts, agentic cost protection, intelligent caching, and compliance-ready audit logging.

Official SDK ✅

pip install aisg — live on PyPI. Native Python client with typed metadata, structured errors, and model discovery.

Recursive Loop Protection ✅

Shipped. Detects and kills runaway agent loops before they drain your credits. 60s window, 30s cooldown.

Semantic Caching ✅

Shipped. Cache identical DLP-cleaned prompts. Cache hits eliminate the LLM call entirely — zero cost, zero latency. Backed by a low-latency distributed cache.

Webhook Notifications ✅

Shipped. HMAC-signed webhooks for PII blocks, prompt injection, redaction, budget alerts, and loop detection. Up to 5 per project.

EU AI Act Logging ✅

Shipped. Hash-chained, tamper-evident audit records with input/output fingerprints, JSONL export, and chain verification API. Ready for August 2026 enforcement.

  • ✅ Official Python SDK (pip install aisg) — shipped, live on PyPI
  • ✅ Recursive agent-loop detection and auto-kill before credits drain — shipped
  • ✅ Webhook notifications for DLP violations, prompt injection, budget alerts — shipped
  • ✅ EU AI Act compliance logging — hash-chained append-only audit trails with JSONL export — shipped
  • ✅ Semantic caching for DLP-cleaned prompts — 100% cost savings on cache hits, zero latency — shipped
Phase 2Shipped

Enterprise Deployment & Hybrid VPC

Deploy the AISG proxy inside your own infrastructure. Prompts never leave your network — DLP, PII redaction, and budget enforcement run on-prem. Cloud dashboard manages policies, analytics, and multi-project governance.

Enterprise-Ready from Day One

SaaS convenience meets on-prem data sovereignty. Hybrid VPC, SSO, RBAC, and SIEM connectors — all shipped.

Hybrid VPC Deploy ✅

Shipped. Run the compiled Go proxy in your VPC. Cloud dashboard manages policies; prompts never leave your infrastructure. Docker Compose or Kubernetes, 3 containers, 4GB RAM minimum.

SAML SSO ✅

Shipped. Self-hosted BoxyHQ Jackson — Okta, Azure AD, Google Workspace, any SAML 2.0 IdP. Auto-provisioning with configurable default roles.

SIEM Connectors ✅

Shipped. Stream security events to Splunk HEC, Datadog Logs, or Microsoft Sentinel. Tokens stored in Secrets Manager, outbound HTTPS only.

RBAC & Team Management ✅

Shipped. 4-tier roles (owner, admin, member, viewer), org model, invitations, and per-team policy assignment.

  • Hybrid VPC Deployment — compiled Go proxy in your VPC ✔️
  • Prompts never leave your network — metadata-only cloud telemetry ✔️
  • 30+ PII entity types with local DLP engine (sub-50ms) ✔️
  • Multi-project support — single proxy serves multiple projects via API keys ✔️
  • Per-project monthly budget enforcement (local + cloud) ✔️
  • Cloud dashboard for policy management, violations & analytics ✔️
  • Policy sync every 30s — zero-downtime policy updates ✔️
  • Docker Compose deployment — 3 containers, 4GB RAM minimum ✔️
  • ✅ SAML SSO — Okta, Azure AD, Google Workspace, any SAML 2.0 IdP
  • ✅ RBAC & Team Management — 4-tier roles, org model, invitations
  • ✅ SIEM connectors — Splunk HEC, Datadog Logs, Microsoft Sentinel
  • Kubernetes deployment support — ready-to-apply manifests with NetworkPolicy ✔️
Phase 3Beta

Agentic AI Governance & MCP Gateway

Our top priority — now in public beta. As teams move from chatbots to autonomous agents, governance shifts from static prompt filtering to real-time control of the tool plane. The MCP Gateway extends AISG’s inline trust-boundary model to the tool↔LLM flow — the same DLP and injection defense, applied where agents call tools.

Governing Autonomous AI

When AI agents call tools, execute code, and make decisions autonomously — who’s watching the tool plane?

MCP Gateway ✅ (Beta)

Live in beta. An aggregating proxy between agents and MCP servers — enforce policy on every tool call and result, across any LLM, in cloud or in-VPC.

Tool-Result DLP ✅ (Beta)

Live in beta. Scan tool and database outputs for PII before they enter the model’s context. Redact or block — fail closed.

Schema Pinning / Rug-Pull Guard

Building. Pin approved tool schemas and block silent redefinition — a known MCP attack vector.

Human-in-the-Loop

Building. Approve or deny high-risk agent actions in real time, with full audit + SIEM streaming.

  • ✅ MCP Gateway — aggregating proxy between agents and MCP servers (streamable HTTP; stdio in Hybrid VPC) — beta
  • ✅ Tool-description poisoning scan — catalog-time injection detection before a tool reaches your agent — beta
  • ✅ Tool-result DLP — redact or block PII in tool outputs before they reach the model — beta
  • ✅ Tool-call argument DLP — stop data exfiltration via tool inputs — beta
  • ✅ Per-direction DLP actions — request off/block, response off/redact/block — beta
  • ✅ Tool allow/deny lists — default-deny allowlists; denylists always win — beta
  • ✅ Graceful degradation + audit/SIEM events on every block / redact — beta
  • ✅ Available in Cloud and Hybrid VPC (in-VPC bundle, credentials never leave your network) — beta
  • Tool-schema pinning & rug-pull detection — block silent tool redefinition (building)
  • Human-in-the-loop approval hooks for high-risk agent actions (building)
  • Behavioral monitoring — loop, exfiltration & scope-creep anomaly detection (building)
Phase 4Building

Cost Intelligence & Advanced Detection

Cut AI Spend Without Sacrificing Security

Granular budget controls and next-generation prompt injection defense.

Multi-Provider Quotas

Set one budget across all providers. Get alerted at 50%, 80%, and 100% via Webhook or Slack.

ML Jailbreak Detection

Deep learning classifiers that catch attacks regex can’t — semantic similarity, encoding exploits, and novel patterns.

Policy-Based Routing

Automatically restrict to low-cost models when a project crosses a spending threshold.

  • Per-project monthly budget hard-stops (hybrid + cloud BYOK) ✔️
  • Multi-provider spending quotas with threshold alerts (50%, 80%, 100%)
  • ML-based jailbreak classifiers beyond regex heuristics
  • Policy-based ‘Budget Mode’ routing to low-cost models when spending thresholds are crossed
Available Now

The AI Intelligence Suite

Stop guessing which model is best for your specific prompts. The world's first Financial & Quality Optimizer for production AI is here.

Smart Model Selection

Automatically pick the best model for each prompt based on quality, latency, and cost.

Cost Optimization

Reduce redundant AI spend through semantic caching and smart routing on production workloads.

Quality Benchmarking

Continuous evaluation of model responses against your quality criteria in real time.

Get notified when new features launch:

Get 1,000,000 Free Credits

No credit card required · Start in 60 seconds