apex_ai

Apex AI — Multi-Provider AI Orchestration for Drupal

Apex AI bridges 9 AI providers into Drupal's AI ecosystem through the MonkeysLegion Apex orchestration engine. One provider plugin — every model, every middleware, zero vendor lock-in.

Why Apex AI?

Drupal's AI module gives you a provider API. Apex AI gives you an AI operations platform:

9 providers, 24+ models — Anthropic, OpenAI, Google, xAI, Cohere, DeepSeek, Groq, Mistral, Ollama
Smart routing — Auto-select models by cost, quality, or latency
Middleware pipeline — Retry, rate limiting, caching, guardrails, telemetry
Fallback chains — Automatic failover when a provider goes down
Full observability — Cost dashboard with filters, pipeline run history, audit logs, Drush CLI
Enterprise guardrails — Token budgets, PII filtering, content moderation
GDPR compliance — Per-user data export and deletion, audit retention policies
Google Service Account support — Auto-exchanges JSON credentials for OAuth2 tokens

Any Drupal module that uses the AI API — CKEditor AI, AI Automators, AI Search, AI Chatbot, AI Translate — works with Apex AI automatically.

What's New in v2.0.0

Feature Description Cost Dashboard Real-time cost tracking with period/provider filters, summary cards, CSV export Run Recording Every AI call records provider, model, tokens, cost, latency, and status Audit Trail Full audit log entity: who triggered, what prompt, which model, what response GDPR Compliance Per-user data export and deletion REST endpoints PII Redaction Auto-redact emails, phones, SSNs, credit cards from audit logs Content Governance AI watermarking, quality scoring, brand voice enforcement Google OAuth2 Auto-detect service account JSON and exchange for Bearer tokens Dual Auth Google provider supports both API keys (AIza...) and OAuth2 tokens (ya29...) HTTP/1.1 Forced Fixes HTTP/2 PROTOCOL_ERROR in Docker/containerized environments Extended Timeout 120s default timeout for all providers (configurable) External Integrations Langfuse, OpenTelemetry, Slack, Microsoft Teams webhook support MCP Server Model Context Protocol server for Claude Desktop, Cursor, and other MCP clients ECA Actions Auto-tag, summarize, alt text, quality score, translate — all no-code Agent Crews Multi-agent team orchestration with role-based collaboration Circuit Breaker Automatic provider disable/recovery after consecutive failures Upgrade Hooks Automatic entity schema creation on module upgrade

Supported Providers & Models

Provider Models Auth Strengths Anthropic Claude Sonnet 4, Opus 4, Haiku 4 API Key Reasoning, safety, long context OpenAI GPT-4.1, Mini, Nano, o3, o4-mini Bearer Broad capability, tool use Google Gemini 2.5 Pro, Flash API Key or Service Account JSON Multimodal, cost-effective xAI Grok 3, Mini, Fast Bearer Real-time knowledge Cohere Command R+, Command R Bearer RAG, enterprise search DeepSeek Chat, Reasoner Bearer Math, code, reasoning Groq LLaMA 3.3 70B, Mixtral 8x7B Bearer Ultra-low latency Mistral Large, Small Bearer European compliance, multilingual Ollama Any local model None Air-gapped, private, zero cost

Smart Routing Modes

Select a routing strategy instead of a specific model:

apex:auto — Intelligent model selection based on prompt analysis
apex:cost — Route to the cheapest available provider
apex:quality — Route to the highest-capability model
apex:latency — Route to the fastest provider

Combined with fallback chains and the circuit breaker, your AI requests survive provider outages automatically.

Cost Dashboard

Enterprise-grade cost monitoring at /admin/config/ai/apex/costs:

Summary cards — All-time cost, today, this month, total requests, input/output tokens, avg latency
Period filter — Today, last 7 days, this month, last 3 months, last year, all time
Provider filter — Drill down to a single provider
Cost breakdown table — Per-provider aggregation of requests, tokens, cost, and latency
Recent runs — Last 50 AI calls with full metadata and status indicators
CSV export — Download all run data for billing reconciliation
Middleware status — Visual overview of active pipeline components
Theme compatible — Works with Claro, Gin, Seven, or any admin theme

Middleware Pipeline

Every request passes through a configurable middleware stack:

Middleware Purpose Config Retry Exponential backoff on transient failures Max attempts, delay (ms) Rate Limit Per-provider request throttling Requests/minute Response Cache TTL-based caching to reduce API costs TTL (seconds) Guardrails Token limits, PII blocking, content moderation Max tokens, filter toggle Telemetry Automatic cost, token, and latency recording Sink (Drupal logger, Langfuse, OTEL)

Observability & Governance

Audit Trail

Full audit log entity: user, prompt, response, model, tokens, cost
PII redaction (email, phone, SSN, credit card — configurable)
Retention policy with auto-purge after N days
Admin review at /admin/config/ai/apex/audit

GDPR Compliance

GET /api/apex-ai/gdpr/export/{uid} — Export all AI interactions for a user
DELETE /api/apex-ai/gdpr/delete/{uid} — Permanently erase a user's AI data

Content Governance

AI-generated content watermarking (metadata flag)
Quality scoring for AI outputs
Brand voice enforcement via guardrail rules

External Integrations

Langfuse — LLM observability platform
OpenTelemetry — Distributed tracing
Slack / Microsoft Teams — Budget alert notifications

Submodules

Apex AI Orchestration

Pipeline and crew management with a full cost dashboard.

Named AI workflows (pipelines) with configurable models
Multi-agent crew collaboration
Content safety guardrails (PII filtering, token budgets)
Cost Dashboard — Real-time per-provider cost tracking with filters
Pipeline Runs — Full execution history with status and metrics
Audit Logs — Complete interaction history with PII redaction

Apex AI Agents

Autonomous AI agents using a ReAct-style reasoning loop.

Config entity-based agent definitions with goals, tools, and memory
Multi-step reasoning with tool calling
5 built-in tools: CreateNode, UpdateField, SearchContent, QueryViews, SendEmail
Async execution via queue workers or webhook triggers
Crew orchestrator for multi-agent team coordination

Apex AI Prompts

Reusable prompt template library.

Config entity for prompt templates with token variables ({{ variable }})
Version tracking and live preview
Categorization and tagging

Apex AI ECA

No-code AI automation via ECA.

AutoTagAction — Auto-tag content using AI classification
GenerateSummaryAction — Generate content summaries on save
GenerateAltTextAction — Generate image alt text
QualityScoreAction — Score content quality
TranslateFieldAction — Translate field values

Apex AI MCP Server

JSON-RPC 2.0 Model Context Protocol server.

Exposes Drupal content as MCP resources
Registers Drupal operations as MCP tools
Compatible with Claude Desktop, Cursor, and other MCP clients

Apex AI Vector Database

Zero-dependency VDB provider for the AI Search module.

SQLite-based vector storage with cosine similarity
Full Search API integration
Automatic content chunking and embedding

REST API

Method Endpoint Description POST /api/apex-ai/generate Generate AI response POST /api/apex-ai/stream Streaming AI response (SSE) GET /api/apex-ai/health Provider health check GET /api/apex-ai/models List available models GET /api/apex-ai/costs Cost summary (JSON) GET /api/apex-ai/costs/export Full run data (CSV) GET /api/apex-ai/gdpr/export/{uid} Export user AI data DELETE /api/apex-ai/gdpr/delete/{uid} Delete user AI data POST /api/apex-ai/search RAG semantic search POST /api/apex-ai/mcp MCP JSON-RPC endpoint

Drush Commands

Command Alias Description drush apex:test <prompt> axt Test a provider with a prompt drush apex:providers axp List all 9 providers with status drush apex:costs axc Cost breakdown by provider drush apex:runs axr Recent pipeline execution log drush apex:agent:run <id> <goal> axag Execute an AI agent

Architecture

Drupal AI Module (CKEditor AI, AI Automators, AI Search...)
          │
          ▼
┌─────────────────────────────────────────────────┐
│            ApexProvider Plugin                   │
│   ChatInterface + EmbeddingsInterface            │
│                                                  │
│  translateInput → generate → recordRun → output  │
└─────────────┬───────────────────────────────────┘
              │
              ▼
┌─────────────────────────────────────────────────┐
│              ApexFactory                         │
│                                                  │
│  resolveProvider → resolveApiKey → createProvider │
│  Google: JSON → JWT → OAuth2 → Bearer Token      │
└─────────────┬───────────────────────────────────┘
              │
              ▼
┌─────────────────────────────────────────────────┐
│       MonkeysLegion Apex SDK (v1.0.4)            │
│                                                  │
│  Anthropic │ OpenAI │ Google │ DeepSeek │ Groq   │
│  Mistral   │ xAI    │ Cohere │ Ollama            │
│                                                  │
│  HTTP/1.1 forced · Dual auth · 120s timeout      │
└─────────────────────────────────────────────────┘

Installation

composer require drupal/apex_ai
drush en apex_ai -y

Quick Start Recipe

php core/scripts/drupal recipe web/modules/custom/apex_ai/recipes/apex_ai_quickstart

Enterprise Recipe

php core/scripts/drupal recipe web/modules/custom/apex_ai/recipes/apex_ai_enterprise

Requirements

Drupal ≥ 11.2
PHP ≥ 8.4
drupal/ai ≥ 1.4
drupal/key ≥ 1.18
monkeyscloud/monkeyslegion-apex ≥ 1.0.4

Docker & Containerized Environments

HTTP/1.1 forced — SDK forces CURL_HTTP_VERSION_1_1 to prevent HTTP/2 stream errors
120s timeout — Extended timeout for slow container networking
OPcache — Restart PHP-FPM after code changes (docker compose restart php)

Security

API keys stored via the Key module — never in config YAML
Google service account JSON exchanged for short-lived OAuth2 tokens (1-hour TTL)
All admin routes require administer apex_ai permission
PII auto-redaction in audit logs
GDPR-compliant per-user data export and deletion

Version	Type	Release date
2.0.0	Stable	May 12, 2026
1.0.0	Stable	May 12, 2026