Agent IAM

OpenClaw runs your agents. Carryall guards the data.

Every agent gets its own cryptographic identity. Every action gets a signed envelope. Every access gets logged. Your auditors will thank you.

The Stack

Execution

OpenClaw / LangChain / AutoGen / Claude Code

runs agents
Authorization

Carryall — Agent IAM

guards data
Your Data

Vaults / Databases / APIs / File Systems

sensitive data

Carryall sits between your execution layer and your data. It doesn't replace your agents — it governs them.

How It Works

1

compile_policy

Describe what the agent needs in plain English. A local LLM determines the minimal permissions required.

2

Signed Envelope

Ed25519 cryptographic envelope. Scoped to specific resources. Time-limited. One envelope per task.

3

check_access

Every read and write is verified against the envelope before execution. No valid envelope, no access.

4

audit_log

Immutable, agent-attributed trail. Every action logged with who, what, when, and why.

The Demo

Two agents. Same machine. Same vault. Different keys.

accountant-agent

Can read tax documents

Denied investment data

Denied health records

investment-agent

Can read portfolio data

Denied tax documents

Denied health records

Both enforced cryptographically. Both logged immutably.

Show this to your compliance officer.

What Makes This Different

NemoClawYAML GuardrailsCarryall
When it operatesDeployment-timeDeployment-timeRuntime, every action
What it controlsNetwork callsAPI accessData access per document
Audit trailNetwork logsNoneAgent-attributed, immutable
Policy formatYAMLYAMLNatural language → LLM → Rego
PlatformNVIDIA hardwareVariesAny hardware, any framework
Policy decisionsCloudCloudLocal LLM — never leaves your network

Want to try it?

Zero to a fully authorized agent stack on your own hardware. Four steps, under an hour.