Privacy Policy

Smara is memory infrastructure for AI agents, operated by Prajana AI (“we”, “us”). This policy explains what we collect, how we use and store it, what we share, how long we keep it, and how to reach us. It covers two kinds of people: the developers and teams who use the Smara portal and API (“customers”), and the end users whose information a customer’s agents may store in Smara as memory (“memory data”).

What we collect

Account data. When you create an account we collect your email address and, if you sign in with Google, the basic profile information Google shares (name and email). Authentication is handled by Firebase Authentication; we never see or store your password.

Tenant and API-key metadata. Tenant names, membership, API-key names, scopes, creation and last-used timestamps. API-key secrets are stored only as SHA-256 hashes — we cannot recover a secret after it is shown once at creation.

Memory data.Whatever a customer’s agents write through the API or MCP tools: conversational turns, episodic events, semantic facts, knowledge-graph entities and relationships, and session scratchpads. This content is under the customer’s control and may include personal information about the customer’s own users; the customer is responsible for having a lawful basis to store it.

Operational logs. Standard request logs (IP address, endpoint, timestamps, status) for security, debugging and abuse prevention.

How we use it

We use data solely to provide the service: authenticating you, enforcing tenant isolation, storing and retrieving memory, running background consolidation (which distills raw events into durable facts and graph structure), computing relevance and decay, and operating and securing the platform. We do not sell data, we do not use memory data for advertising, and we do not use your content to train our own or anyone else’s models.

Where it lives and who processes it

All data is stored on Google Cloud Platform in the United States (region us-central1): Cloud Spanner for durable memory, Firestore for tenant and skill metadata, Memorystore for short-lived working memory, and Cloud Storage for skill definitions. During consolidation and retrieval, memory content is processed by Google’s Vertex AI services (Gemini models for extraction, embedding models for search) under Google Cloud’s data-processing terms, which do not permit Google to train on this data. These Google Cloud services, plus Firebase Authentication, are our subprocessors. We share data with no one else, except where the law requires it.

How long we keep it

Memory is designed to forget. Short-term working memory (recent turns, scratchpads) expires automatically within hours to days. Durable facts and graph entities decay based on importance and usage: items whose effective relevance falls below a floor are first expired (closed, no longer retrievable) and then, after a retention window, permanently deleted. Customers can also explicitly forget any fact or entity at any time from the portal or API; deletions are recorded in an audit trail that captures what was removed and when. Account data is kept while your account is active and deleted on request.

Your choices and rights

You can browse and delete memory from the portal, revoke API keys, and request account deletion or a copy of your data by emailing us. If you are an end user of one of our customers and believe your information is stored in Smara, contact that customer first — they control the tenant that holds it — and we will support their deletion requests. Depending on where you live, you may have additional rights (access, correction, deletion, portability) under laws such as GDPR or CCPA; we honor verified requests to the extent those laws apply.

Security

Traffic is encrypted in transit (TLS) and at rest. API keys are scoped per tenant, carry least-privilege scopes, and are stored only as hashes. Control-plane and data-plane credentials are separate: portal sessions cannot call the memory API and API keys cannot administer tenants.

Changes and contact

We will post any material changes to this page and update the effective date. Questions, requests, or concerns: bharat@prajana.ai. See also our Terms of Service.