AI Agent Assembly is an open-source governance platform for AI agents. It enforces policy, tracks budget, and audits every action your agents take across three independent interception layers — in-process SDKs, a sidecar proxy, and eBPF kernel hooks — so you can ship multi-agent fleets without losing control of what they do.
brew install agent-assembly/tap/aasm# pip
pip install agent-assembly-python
# uv
uv add agent-assembly-python
# poetry
poetry add agent-assembly-python# pnpm (recommended)
pnpm add @agent-assembly/sdk
# npm
npm install @agent-assembly/sdk
# yarn
yarn add @agent-assembly/sdkdocker pull ghcr.io/agent-assembly/python:latestcurl -fsSL https://get.agentassembly.dev | shagent-assemblyis currently in a pre-stable alpha channel.- Homebrew tap has
Formula/aasm.rbwithversion "0.0.1", but checksum placeholders are still all zero, so this should be treated as pre-stable/bootstrap state until real release artifacts and checksums are published.
- Lowest-intrusion integration: one-line init across major agent frameworks.
- Three-layer interception: semantic to protocol to kernel, with selectable depth.
- Secret Injection: real credentials never appear in LLM context windows.
- Tool Execution Sandbox: isolated execution to reduce security risk.
- Human-in-the-loop Gate: high-risk actions require human review.
- Agent Identity and Zero-trust A2A
- Cost and Token Budget Governance
- Org / Team / Agent hierarchy management
- Audit Trail and Compliance Export
- Local Dev Mode (zero-config, OSS)
- Self-hosted Control Plane (enterprise on-prem)
- SaaS Cloud
Prioritize these three above the fold: lowest-intrusion integration, Secret Injection, and Human-in-the-loop Gate.