Secure infrastructure for AI
The trust layer for AI, agents, humans, bots and SI
AzatAI Stack gives AI products what they need to earn trust: sign-in for people and AI agents, a safe home for data, and encrypted file storage. Run it on your own infrastructure, on AzatAI Cloud, or with a partner cloud.
Standards and tools we build on
- Rust
- OpenID Connect
- OAuth 2
- Passkeys
- WebAuthn
- RFC 8693 token exchange
- Model Context Protocol
- Ed25519
- age encryption
- PostgreSQL
- SQLite
- S3-compatible storage
- Docker
- Kubernetes
- Cloudflare Workers
What we do
One stack under every product
We build AzatAI Stack, our own products on it, and secure systems for clients. Every product proves the platform in real use before a client depends on it.
01 · Products
Our products
AI products that AzatAI builds and runs, on the same foundation as everything else.
02 · Client systems
Systems for organizations
Secure AI and software that we build, deploy where the client decides, and keep maintained.
AzatAI Stack
Rust modules · take only what you need
Auth
Available
Data
Available
Objects
Available
Access
Planned
Jobs
Planned
Search
Planned
Export
Planned
Runs where you choose
Your infrastructure
Your own servers or data center
AzatAI Cloud
We run and update it for you
Partner cloud
A cloud provider we work with
Docker · Kubernetes · PostgreSQL · S3-compatible storage · Cloudflare Workers
Identity for AI
Sign-in for humans and machines
AI agents read data, call tools and act for people. AzatAI Auth gives every agent an identity of its own, so every action has an owner.
$ curl -X POST https://auth.example.com/token \ -d grant_type=…:grant-type:token-exchange \ -d subject_token=$USER_ACCESS_TOKEN \ -d subject_token_type=…:token-type:access_token \ -d resource=https://files.example.com/api \ -d client_assertion=$SIGNED_WITH_AGENT_KEY# Claims in the new access token{ "sub": "user_4f9c", "aud": "https://files.example.com/api", "client_id": "research-agent", "act": { "sub": "research-agent" }}Passkeys for people
People sign in with passkeys and security keys. There are no passwords to steal, reuse or phish.
Credentials for agents
Each agent is its own OAuth client and proves who it is with its own key, not with a shared secret.
Acting for a person
With token exchange, an agent can act for a user. The token names both, so you always know who acted for whom.
Ready for MCP
MCP clients connect with standard OAuth and register by URL, through a Client ID Metadata Document.
Security model
Security is the architecture
These rules hold for every product and system we build.
01
Your keys
Every file is encrypted before it reaches storage. The owner holds the keys, so a stolen disk, bucket or backup gives nothing away.
02
Hosting you choose
Run it on your own infrastructure, on AzatAI Cloud, or with a partner cloud. Self-hosting is always an option.
03
Memory-safe code
Our services are written in Rust, which rules out whole classes of memory bugs before the code ships.
04
No telemetry
Our software sends no usage data to us or to anyone else. What people and agents do stays with you.
05
Proven cryptography
We use well-known, publicly reviewed cryptography such as age and Ed25519. We never invent our own.
06
No lock-in
Stored files stay readable with the standard age tool, and data lives in PostgreSQL or SQLite. You can leave and keep everything.
Build AI that people can trust
Write to us in general terms. If your work is sensitive, we agree on a safer channel before you share any details.