# aamp

The self-hosted control layer for AI agents. Runs on your infrastructure.

## What it is
aamp runs AI agents on hardware you control while keeping credentials, policy, cost control and audit host-side. The model gets tools; the host keeps authority. It runs as a single Go binary against a local SQLite database.

## Core capabilities
- Governance: 35-permission RBAC with per-resource grants, inbound/outbound LLM safety firewalls, fourteen audit log domains
- Agent/Ingress separation: behaviour (the Agent) and exposure (the Ingress) are separate objects, separately audited
- Cost governance: tiered routing to smaller models, per-agent loop budgets, reasoning-effort caps, per-token attribution
- Model freedom: any OpenAI-compatible provider, local open weights or hosted frontier, bound per agent, with fallback routing designed in
- Retrieval: host-side document ingestion and retrieval, with per-query usage logging
- Deterministic Starlark workflows that invoke models as bounded, audited steps

## Positioning
Hyperscalers give you isolation inside their cloud. Self-hosted automation platforms give you your own infrastructure without the governance layer. aamp is the intersection: a full permission and audit layer, entirely on hardware you control.

## Requirements
Linux host with KVM; a 32-byte secrets master key. Single Go binary plus SQLite; backup is a file and two directories.

## Not the right tool when
You want a Slack-to-Sheets automation this afternoon, need thousands of pre-built connectors, or have no one who can run infrastructure.

## Key claims
- Runs on your infrastructure, in your jurisdiction — no third-party control plane
- Model-agnostic: OpenAI-compatible, local or hosted
- Cost stays flat as you automate more
- Every action (prompt, retrieval, tool call, output, token spend) is logged host-side
- Provides EU AI Act evidence (lineage, attribution, audit); compliance itself remains an organisational obligation

## Pages
- / — overview
- /product/ — architecture and capabilities
- /how-we-differ/ — comparison vs n8n, make.com and direct LLM APIs
- /docs/ — documentation
- /get-started/ — try it free, install it yourself, or talk to us

## Contact
https://getaamp.com/get-started/