Your agents make API calls.
They should never see your keys.

Clauth is a local daemon that sits between your AI agents and the APIs they call. It holds your credentials in an encrypted vault and injects them at request time — so agents get the job done without ever touching a secret.

Encrypted vault (AES-256-GCM) Per-agent scope enforcement Behavioral anomaly detection Tamper-evident audit log

Up and running in 60 seconds

Install, store your credentials once, and every agent you run is automatically protected.

1

Install & start the daemon

$ npm install -g clauth-ai
$ clauth init
$ export CLAUTH_ADMIN_TOKEN='set-admin-token'
$ clauth daemon

One-time setup. Runs locally on your machine — nothing leaves your network.

2

Store your credentials

$ export GITHUB_PAT='ghp_xxx'
$ export OPENAI_API_KEY='sk-xxx'
$ clauth store --handle github-main --provider github --secret-env GITHUB_PAT
$ clauth store --handle openai-main --provider openai --secret-env OPENAI_API_KEY

Encrypted at rest with AES-256-GCM. You set the passphrase, nobody else.

3

Grant scoped access to skills

$ clauth grant --skill my-agent --provider github --scope github:read --rpm 60
$ clauth grant --skill my-agent --provider openai --scope openai:chat --rpm 60
$ clauth grant --skill my-agent --provider stripe --scope stripe:charges --rpm 30

Each agent gets only the permissions it needs. Nothing more.

Operator quick start

Copy/paste runbook to provision a skill and validate a real brokered call.

$ export CLAUTH_PASSPHRASE='your-long-passphrase-here'
$ export CLAUTH_ADMIN_TOKEN='set-admin-token'
$ clauth init
$ clauth daemon

# in another terminal
$ export CLAUTH_PASSPHRASE='your-long-passphrase-here'
$ export GITHUB_PAT='ghp_xxx'
$ clauth store --handle github-main --provider github --secret-env GITHUB_PAT
$ clauth grant --skill my-agent --provider github --scope github:read --rpm 60
$ clauth skill-token issue --skill my-agent
$ export CLAUTH_SKILL_TOKEN='<issued-token>'

$ curl -sS -X POST http://127.0.0.1:4317/clauth/v1/proxy \
  -H "content-type: application/json" \
  -H "x-clauth-skill-token: $CLAUTH_SKILL_TOKEN" \
  -d '{ "provider":"github","credentialHandle":"github-main","scope":"github:read","method":"GET","endpoint":"https://api.github.com/user" }'

$ curl -sS http://127.0.0.1:4317/clauth/v1/status

Expected: proxy returns a 200 wrapper with upstream payload; status shows auditIntegrity.valid true; audit log contains proxy.allow.

For skill developers

Your skill never sees a credential. Just tell Clauth what you need — it handles the rest.

my-skill.ts
import { ClauthClient } from "clauth-ai/client";

const clauth = new ClauthClient({
  skillId: "my-agent",
  skillToken: process.env.CLAUTH_SKILL_TOKEN
});

// Make an API call — Clauth injects credentials
const repos = await clauth.fetch(
  "github",
  "github-main",
  "github:read",
  "https://api.github.com/user/repos"
);

console.log(repos.body); // Your repos. No token in sight.

Zero credential exposure

Your code never handles secrets. Clauth injects auth headers at the network boundary.

Scoped by default

Request only what you need. Clauth enforces least-privilege — unauthorized calls are blocked and logged.

One line integration

Drop in the SDK. Built-in policies for popular providers; custom providers work with operator-defined allowlists.

Behavioral protection

Clauth watches for anomalies — unusual endpoints, rate spikes, off-hours activity — and blocks them automatically.

Built for trust

Clauth is designed so you don't have to trust Clauth. Everything is local, auditable, and under your control.

Runs on your machine

No cloud service, no SaaS, no third-party credential storage. Clauth is a local daemon — your secrets stay encrypted on disk and are only sent to the provider APIs you call.

Zero runtime deps

No third-party runtime packages. Clauth runs on Node.js built-in modules; dev deps are TypeScript tooling only.

Tamper-evident audit log

Every request, every grant, every anomaly — hash-chained and append-only. If someone edits history, integrity verification fails.

Open source

MIT licensed. Audit the source, fork it, self-host it. Security through transparency, not obscurity.

Emergency kill switch

One command revokes every grant instantly. Compromised skill? Shut it down in seconds, not hours.

Security advisory monitoring

Clauth polls public advisory feeds and auto-revokes affected credentials before you even hear about the breach.

Under the hood

Explore the architecture, see a request flow in action, or compare the before and after.

OpenClaw Today
Lose Your Identity

A malicious skill steals your OAuth tokens and impersonates you across every connected platform

Lose Your Funds

Payment credentials sitting in plaintext — one bad install drains your accounts

Credit Card Stolen

Stored card details are readable by any skill with file access. No encryption, no barrier.

API Bill From the Dark Web

Your cloud API keys get exfiltrated and sold — strangers run up thousands on your tab

No Access Control

Every skill has god-mode access to every credential you own

No Detection

Zero anomaly monitoring — you find out when the bill arrives or the damage is done

No Audit Trail

No record of what was accessed, when, or by whom

No Recovery Plan

When it happens, you manually rotate every key and hope you got them all

With Clauth
Secret Storage

AES-256-GCM encrypted vault

Skill Access

Skills receive scoped handles, never tokens

Scope Control

Granular provider:action grants

Anomaly Detection

Per-skill baselines with critical blocking

Audit Log

Hash-chained append-only event stream

Breach Response

Emergency revoke and advisory-driven revocations

Identity Verification

Multiple private verification options

Operational Friction

Near-zero during normal operation

Attack Scenario: Malicious Skill Installation

Without Clauth

Skill reads plaintext credential files, exfiltrates keys, and pivots across all connected providers before the user notices.

With Clauth

Skill has no secret file access. Unauthorized scope request is denied, logged, and surfaced as an anomaly with immediate revoke options.

If the interactive deep dive doesn't load, scripts may be blocked. The default view above is still accurate.

Documentation

Everything you need to get started, integrate your skills, and harden your setup.