Enterprise Agent Operating System
17-plane kernel, 16 platform adapters, 1,584 tests, 48/48 protocol conformance. Build, deploy, and govern autonomous AI agents at production scale.
Platform Overview
Identity → Safety & Alignment bounded planes
OpenClaw, CrewAI, LangGraph, Claude Code, MCP, A2A + more
Multi-model routing with cost/latency optimization
Full protocol compliance across all adapters
Enterprise-grade test coverage across all modules
OWASP-compliant, zero critical vulnerabilities
Enterprise Features
Executive Dashboard
Real-time agent orchestration with role-based views. Monitor fleet health, budgets, and compliance from a single pane.
Agent Registry
Discover, install, and manage certified agents. 8 marketplace agents with digital signing and lifecycle management.
Skill Registry
Versioned, composable agent skills. Swappable behavioral units with dependency resolution and provenance tracking.
Live Agent Execution
Execute agents in real-time with streaming output. HITL approval gates, retry with backoff, and Saga compensation.
Memory Visualization
4 memory types visualized: semantic, procedural, episodic, organizational. Knowledge graph with 142 entities and 163 edges.
Workflow Builder
DAG-based workflow composition. Visual task chaining with checkpointing, rollback, and conditional branching.
Event Timeline
Immutable event sourcing with full replay capability. Every execution produces a verifiable receipt.
Evaluation Dashboard
14-dimension agent scoring with historical trends. Regression detection and continuous improvement tracking.
LLM Router
Intelligent model selection across 10 providers. Cost/latency/quality optimization with fallback chains.
MCP Tool Registry
Model Context Protocol integration. Auto-discover and register external tools with capability-based permissions.
Runtime Metrics
OpenTelemetry traces, cost tracking, SLO dashboards. Full observability across the agent mesh.
Enterprise Demo Mode
Pre-built demo scenarios: multi-agent workflow, executive planning, memory replay, event sourcing, skill execution.
Architecture
Dual-Runtime Engine
Python Enterprise Kernel — 17 bounded control planes, 16 adapters, governance, orchestration, certification, evaluation, digital twin, economics.
TypeScript Monorepo — Express API (33 endpoints), React dashboard, CLI, MCP server, 82 source modules across 28 domains.
Shared Layer — Supabase (PostgreSQL), event bus, memory persistence, skill registry, cron scheduler.
Governance Pipeline
| Gate | Description |
|---|---|
| 1. Intent | Parse and validate agent objective |
| 2. Risk | Classify risk level (LOW→CRITICAL) |
| 3. Policy | Match against allowed capabilities |
| 4. Approval | Route to governance board if required |
| 5. Execution | Dispatch through capability router |
| 6. Receipt | Immutable execution record |
Model Suite
AIF-Kernel
Core enterprise agent runtime. 17 control planes, event sourcing, CQRS command bus. The central nervous system of AIF.
View Model →AIF-Orchestrator
Multi-agent coordination engine. DAG workflow execution, retry/rollback, HITL gates, and Saga compensation patterns.
View Model →AIF-Planner
Task decomposition and dependency resolution. Converts objectives into executable DAGs with resource estimation.
View Model →AIF-Execution
Capability router and execution engine. Dispatches tasks to the optimal adapter with cost/latency awareness.
View Model →AIF-Router
Intelligent model selection across 10+ providers. Quality-aware routing with automatic fallback chains.
View Model →Dataset Ecosystem
aif-system-prompts
Versioned system prompts, policy definitions, and evaluation criteria for enterprise agents.
View →aif-agent-specifications
Structured agent definitions, capability manifests, and skill contracts.
View →aif-benchmarks
Evaluation datasets for agent performance: GAIA, SWE-bench, AgentBench results.
View →aif-agent-skills
Reusable agent skill definitions with versioning, dependencies, and provenance.
View →aif-execution-events
Immutable execution event logs for audit replay and compliance verification.
View →Performance & Benchmarks
Get Started
Clone the repo, install dependencies, and run your first agent in under 5 minutes.
GitHub Repository Hugging Face Hub