KavachOS is open source. Cloud launching soon.
kavachOS

FAQ/FREQUENTLY ASKED

Answers to thequestions we hear most.

Everything you need to know about kavachOS. Can't find it here? Ask on GitHub Discussions or email support@kavachos.com.

01/GENERAL
  • What is kavachOS?

    kavachOS is an open source TypeScript auth library that handles both human authentication and AI agent identity. It ships scoped permissions, delegation chains, audit trails, and MCP OAuth 2.1 in one package, so you do not need to bolt on separate services for human login and agent access.

  • How is kavachOS different from Auth0 or Clerk?

    Auth0 and Clerk solve human sign-in. kavachOS does that too, but also gives every AI agent its own cryptographic bearer token, scoped permissions, and a full audit trail. No other auth library ships agent identity as a built-in feature rather than a workaround using service accounts or bare API keys.

  • Is kavachOS free?

    The open source library is MIT licensed and free forever. kavachOS Cloud (managed hosting) has a free tier for up to 1,000 MAU with no credit card required. Paid plans start at $29 per month.

  • What runtimes and frameworks does kavachOS support?

    kavachOS runs on Node.js, Bun, Deno, and Cloudflare Workers. There are adapters for Next.js, Hono, Express, Fastify, Nuxt, SvelteKit, Astro, NestJS, SolidStart, and TanStack Start. Client libraries exist for React, Vue, Svelte, Expo, and Electron.

  • What databases does kavachOS support?

    SQLite, PostgreSQL, and Cloudflare D1 are supported today. MySQL and MongoDB adapters are on the roadmap. The database layer uses Drizzle ORM, so adding a new adapter is straightforward if your database is not yet listed.

02/AGENT IDENTITY
  • What is an agent identity?

    An agent identity is a first-class principal in kavachOS, the same way a user is a principal. Each agent gets its own cryptographic bearer token, a set of scoped permissions, a TTL, and an independent audit trail. You can create, rotate, and revoke an agent identity without touching the user account it was delegated from.

  • How do delegation chains work?

    An agent can create sub-agents and grant them a strict subset of its own permissions via kavachos.agents.delegate(). If the parent agent is revoked, all downstream agents in the chain immediately lose access. You can set a maximum chain depth to prevent runaway delegation. Every step in the chain is logged to the audit trail with the full ancestry visible.

  • What happens if an agent token leaks?

    Revoke the token via the dashboard or the kavachos.agents.revoke() API. Revocation is instant: the token is invalidated in the KV store and all downstream delegates created from it are invalidated at the same time. Because agents have scoped permissions rather than root access, blast radius is limited to the specific resources the token could reach.

  • Do agents count toward my Monthly Active User (MAU) total?

    No. Agents are billed separately from human users. The free tier includes unlimited agents. On paid plans, agents are metered by action count rather than by identity count, so you are not penalized for creating many short-lived or task-specific agents.

03/MCP OAUTH
  • What is MCP OAuth 2.1?

    The Model Context Protocol (MCP) defines how AI agents interact with tools. kavachOS ships a spec-compliant OAuth 2.1 authorization server for MCP, implementing PKCE S256, dynamic client registration (RFC 7591), resource indicators (RFC 8707), and server metadata (RFC 8414). Any MCP client that follows the spec can authenticate against your kavachOS endpoints without custom code.

  • Do I need kavachOS to use MCP?

    No. MCP OAuth is an optional plugin you enable if your agents use the Model Context Protocol. kavachOS works without it for standard auth and agent identity. You can add MCP OAuth later without migrating your existing users or agents.

  • Which MCP clients are supported?

    Any client that implements the MCP specification, including Claude Desktop, Cursor, Windsurf, and custom clients built with the MCP TypeScript SDK or Python SDK. kavachOS does not hard-code client allowlists because it uses dynamic client registration per RFC 7591.

  • How is PKCE handled?

    kavachOS enforces PKCE S256 for all authorization code flows. The code verifier is never transmitted to the authorization server; only the S256 challenge is sent at the start and the plain verifier is verified locally at token exchange time. Public clients (browser apps, CLI tools, agent runtimes) are supported without client secrets.

04/PRICING AND PLANS
  • What is the free tier?

    The free tier supports up to 1,000 MAU, unlimited agents, seven days of audit history, all auth methods, agent identity, and MCP OAuth. No credit card is required. It is intentionally generous enough to ship a real side project or validate a startup idea.

  • What counts as a Monthly Active User?

    A human user who authenticates at least once in a calendar month. Agents do not count as MAU. Read-only operations like token verification do not count unless they involve issuing a new session.

  • Is there an enterprise plan?

    Yes. Enterprise includes custom MAU limits, SLA-backed uptime, SAML for dashboard SSO, SCIM provisioning, compliance reports, and a dedicated support channel. Email support@kavachos.com to discuss requirements.

  • Do you charge per agent?

    Not per identity. Agents are metered by action count on paid plans. The free tier and Starter plan include generous action quotas that cover most small-to-medium deployments. The pricing page has a breakdown by plan.

05/SELF-HOSTING
  • Can I self-host kavachOS?

    Yes. The kavachos npm package is MIT licensed and designed to be self-hosted. Install it, point it at SQLite or PostgreSQL, mount it on your framework adapter, and you are running the full feature set with no dependency on kavachOS Cloud.

  • What infrastructure do I need?

    For a typical Node.js deployment: a PostgreSQL or SQLite database, a KV store for token revocation (Redis, Cloudflare KV, or a database-backed adapter), and a runtime that can handle HTTP. On Cloudflare Workers you need only D1 and KV -- no additional services.

  • How do I migrate from Cloud to self-hosted?

    Export your data from the kavachOS Cloud dashboard (Settings > Export). The export includes users, agents, sessions, and audit events in a portable JSON format. Import it with the kavachos-migrate CLI tool. The migration guide in the docs covers the step-by-step process including DNS cutover.

  • What database does the Cloud platform use?

    kavachOS Cloud runs on Cloudflare D1 (SQLite-compatible) per tenant with Cloudflare KV for session revocation. Each tenant gets an isolated database -- your user data never shares a table with another customer.

06/SECURITY AND COMPLIANCE
  • Are you SOC 2 compliant?

    kavachOS Cloud is in the process of completing its SOC 2 Type II audit. The kavachOS library includes compliance report generation for SOC 2, EU AI Act, NIST AI RMF, and ISO 42001 -- so if you self-host, you can generate evidence for your own audit directly from the platform.

  • How do audit trails work?

    Every auth action, agent operation, and permission change is logged with a timestamp, the acting principal, the full delegation chain, and the outcome. Audit logs are tamper-evident and exported in a structured JSON format that maps to common SIEM ingestion formats. The free tier retains seven days; paid plans retain 90 days.

  • Is kavachOS GDPR-compliant?

    kavachOS Cloud processes data in the EU by default. We are a data processor and provide a standard Data Processing Agreement (DPA) on request. The platform supports user data export and deletion workflows that satisfy GDPR Article 17 right-to-erasure requirements.

  • Does kavachOS address EU AI Act Article 13 transparency requirements?

    Yes. The agent identity model, audit trails, and delegation chain visibility are designed to satisfy Article 13 requirements for transparency and human oversight of AI systems. The compliance report generator produces documentation in the format recommended by the AI Office guidance.

  • What security controls does kavachOS include out of the box?

    Rate limiting per identity (not per IP), HIBP breach checking on password creation, CSRF protection, httpOnly cookies, email enumeration prevention, signed reset tokens with short TTLs, session freshness enforcement, and PKCE for all OAuth flows. The core library ships with over 1,900 tests.

Still stuck?

We answer every question.

Open a discussion or email us. Most replies come back within a day.