Intelligence that Lives Inside Your Data.
Generic AI assistants make you copy-paste context. AeionClaw is the intelligence layer native to Aeion OS — it reads your live database, executes automated reactions to platform events, remembers per-user context across sessions, and manages your business via 89 native MCP tools and 47 module-specific skills.
Your Autonomous Operating Partner
Event-Driven Reactions
Don't wait for a prompt. The Reaction Engine monitors the platform Event Bus. When a high-value order hits or an urgent ticket arrives, AeionClaw automatically runs commands, triages data, and notifies admins via Slack.
Persistent Smart Memory
Extracts facts, preferences, and context from every conversation, storing semantic embeddings in your private, per-tenant memory store. Remembers what you discussed last month without 1M-token context windows.
89 Native MCP Tools + 47 Skills
Because AeionClaw sits on top of your database, it can *do* things. Ask it to "update order #1234 to shipped" or "show all high-priority tickets," and it executes native CRUD securely. Every module ships its own expert skill.
8 Domain-Specific Personas
Auto-routed by context. Order question → Commerce Agent. Deal question → CRM Agent. Ticket question → Support Agent. Each persona has scoped tool access — never wastes tokens loading the wrong context.
Spend Governance
Per-tenant daily/weekly/monthly cost caps. Alert at 80%; at the cap, your configured action fires — warn, soft-block, or hard-block. Every AI call logged with full cost + provider + token breakdown.
Native Compliance
Role-based scope limits enforced at every tool call. Immutable AI action audit log meeting SOC 2 + HIPAA + ISO 27001. Data never leaves your tenant.
What Makes AeionClaw Different
Lives in your database
Every conversation, memory, skill, action log = a row in your tenant DB. No external SaaS holds your AI history.
Same RBAC as everything else
AeionClaw can only do what the user can do. A "Viewer" role cannot trick the AI into running "Delete."
47 module-specific skills
Each module ships its own AeionClaw skill — a packaged, callable expert. See the full skills catalog.
9 supported LLM providers
OpenAI · Anthropic · Google · Azure · Groq · DeepSeek · Ollama · HuggingFace · Cohere. Switch one config, all skills work. See providers catalog.
Frequently Asked Questions
ChatGPT lives outside your data. You paste context in, you get text out, you copy text back into your systems. AeionClaw lives inside your data — it queries your CRM, places orders, files tickets, runs reports, schedules calls *by directly invoking the same MCP actions your APIs expose*. Same RBAC; same audit trail; same data sovereignty.
Yes — when explicitly granted permission. Per-persona scoping limits which actions a Claw conversation can invoke; user-role limits cascade down (an Editor's Claw can't do what only an Admin's Claw can do). High-risk actions (mass delete, customer data export) configurable to require explicit approval click-through.
Each persona is scoped to a domain (Commerce-Ops, Support-Agent, Finance-Manager, etc.) with focused tool access. Incoming user requests are classified to a persona via the router; the conversation locks to that persona's tool surface. Saves tokens (don't load 89 tools into every prompt), improves accuracy (LLM isn't distracted by irrelevant actions), keeps audit clean (no "why did Claw try to refund an order during a fleet conversation?").
Per-tenant rules subscribe to platform events and call a Claw skill to react. "When `helpdesk.ticket.assigned` fires for a Pro-tier customer, call `aeion-helpdesk-agent` to draft a personalized response." The reaction runs autonomously; output goes to a review queue or auto-sends based on policy. Per-event execution budget caps prevent runaway reactions.
9 — OpenAI · Anthropic · Google · Azure · Groq · DeepSeek · Ollama · HuggingFace · Cohere. Switch the provider per persona or per tenant. The platform AI gateway handles cost-tracking + cap enforcement across all of them, so a Claw conversation can't blow the AI budget without alerting the admin.
Yes — the Skill Factory walks you through authoring a custom skill with system prompt, tool access scope, example prompts, and guardrails. Test in sandbox tenant; admin approval before global rollout; version-bumped deploys.