Security
How Tuitix Studio keeps your data and agents safe
AI agents call real tools with real credentials. Tuitix Studio is designed so that secrets never reach the language model, every tenant's data stays separate, and people stay in control of sensitive actions.
Credentials never enter the prompt
- LLM keys, OAuth tokens and API secrets are stored in the Vault, encrypted at rest with Fernet (AES encryption with HMAC authentication).
- Secrets are decrypted only at the moment a tool makes its network call. They are not placed in agent state, prompts or execution history.
- A secret-stripping layer scrubs credential-like values from anything sent to a model, and tool argument schemas shown to the model never include secret fields.
Tenant isolation and access control
- Workflows, chat threads, knowledge bases, credentials and usage are scoped to each account.
- Role-based access: Admins design and publish agents, Consumers chat with the agents shared with them, and platform administrators manage users.
- Workflows can be shared with specific users without exposing their credentials.
Accounts and sessions
- Passwords are hashed with bcrypt; they are never stored in plain text.
- New accounts are verified with a one-time code sent by WhatsApp or email.
- Sessions use signed tokens that expire.
People stay in control
- Human-in-the-loop: require approval before an agent runs selected tools, or let it ask a person when it is unsure.
- Guardrails on input and output: length limits, blocked topics and patterns, a safe fallback reply, and redaction of emails, phone numbers and secret-looking tokens.
- An audit log records configuration changes and sign-ins, with per-user activity views for administrators.
Your infrastructure, your models
- Tuitix Studio can be self-hosted. Data lives in your own PostgreSQL database, including document embeddings (pgvector) — no third-party vector service is required.
- Choose the model provider per agent step: OpenAI, Anthropic, Google, Groq, AWS Bedrock, Azure OpenAI or a local Ollama model.
- Token usage is recorded after every model call, so spend is always visible.
Questions
Evaluating Tuitix Studio for your organisation? Read the getting-started guide or create a free account to try it.