A zero-knowledge encrypted vault on your hardware, under your control — with a sovereign AI assistant that searches, reads, and writes against the vault without ever breaking the seal.
Not another AI chatbot. A sovereign vault built around one structural guarantee: your data is encrypted with keys only you hold — and everything works against that vault without ever breaking the seal.
Your notes are encrypted with AES-256-GCM and keys only you hold. The server stores ciphertext. We cannot read your data even if we wanted to. Auto-lock, trash with restore, encrypted backups, recovery key. This is architecture, not policy.
Chat with your vault. The assistant searches, reads, and writes your notes, runs tools, and streams answers — powered by llama.cpp on your own machine. Your words never leave your hardware.
Chat, storage, and plans connect as MCP services — open protocol, swappable, yours. End-to-end encrypted messaging. Client-side encrypted file storage. Plans that stay yours. More services can come; the assistant stays lean.
Any OpenAI-compatible local model, swapped without losing a byte of your data. The model is the voice — your vault is the constant. Confidential-cloud inference arrives later through zero-trust connectors. Never plain cloud APIs.
A progressive web app and local services on your hardware. No cloud required. Full offline capability. Your vault, your machine, your rules.
Every object stored with the hosted Vault API receives a signed proof certificate, verifiable offline against a published public key. Don't trust the service — verify it. No opaque model weights making decisions about your life.
A new approach to personal AI — trustworthy, user-owned, and built around the vault you already have.
| Traditional Approach | KalaBodha | |
|---|---|---|
| Data sovereignty | Cloud-hosted data | Encrypted vault — keys only you hold. |
| Intelligence | Model-centric intelligence | Vault-centric — your notes, your context, your keys |
| Memory | Session-based context | Your vault compounds — notes, chats, plans persist and stay yours |
| Model | Fixed model choice | Any OpenAI-compatible local model. Swappable. |
| Cloud | Required | Optional — local first; confidential cloud later, zero-trust only |
| Offline | Cloud required | Full offline capability with a local model |
| Your data trains their model? | Varies by provider | Never — by architecture. Nothing leaves your machine. |
| Verifiability | Trust the provider | Published proof key + public audit chain. Check it yourself. |
| Cost | Subscription model | Free. Your hardware, your compute. |
The vault is the product. The assistant works against it — on your hardware. Services connect over MCP. Models are interchangeable voices; your data is the constant.
No vapor — every claim below is true of the product as it stands today.
Encrypted vault, local assistant, and connected services — built, tested, and released (release gate, 2026-10-24).
Public launch of the app — then confidential-cloud inference through zero-trust connectors, without surrendering the vault.
More connected services over MCP. The assistant stays lean; the ecosystem grows.
The vault, the local assistant, and the connected services are built and released — KalaBodha graduated its release gate on 2026-10-24. The hosted Vault API is live for developers today: every stored object returns a signed proof certificate, and its proof key and audit chain are public, so you can verify storage proofs yourself, without trusting the server. The app itself opens to the public at launch — until then it stays behind early access, by design, and collaborator onboarding opens with the launch.
When onboarding opens, these are the areas:
Assistant UI, vault hardening, chat experience, mobile web
New connected services over the Model Context Protocol
Model evals, packaging, LM Studio flows, performance on consumer hardware
Prompt-injection defense, zero-knowledge review, protocol scrutiny
Architecture docs, setup guides, API docs, tutorials, onboarding content
Interface design for everyday people, onboarding flow, brand
Developers: the Vault API is live — see pricing below. Everyone else: request early access to the app, or read the whitepaper.
Non-negotiable. Structural, not policy.
KalaBodha is released. The Vault API is live for developers today; the app opens at public launch. Here is what you can do today:
The Vault API publishes its signing key and its audit chain. Check them yourself — no account needed.
Tell us what you would use KalaBodha for. Invitations go out as launch onboarding opens.
The KalaBodha app is free — your hardware, your compute, AGPL-3.0. What you pay for is the hosted Vault API: client-encrypted storage where every object returns a signed proof certificate, verifiable offline, backed by a public audit chain.
US$5/month
1,000 stored objects · 1 GB. Proof certificate on every object, MCP endpoint, public audit chain.
US$19/month
50,000 stored objects · 100 GB. Everything in Starter, at working scale.
US$99/month
1,000,000 stored objects · 2 TB. Everything in Pro, for fleets of agents and applications.
Checkout is open — subscribe with Stripe: Vault Starter — US$5/month · Vault Pro — US$19/month · Vault Scale — US$99/month. Already hold a Vault API key? Enter its ID at checkout and your plan activates the moment payment lands. New to the Vault API? Your key is issued against your purchase — write to us after checkout and it is bound the same day. Keys are issued per namespace; your content stays encrypted with keys only you hold.