What is Pramiti?
The trust layer between AI agents and enterprise data
Pramiti is the trust layer between AI agents and enterprise data. It answers two questions every enterprise must resolve before giving AI agents access to production systems: "What does this data mean?" and "Is this action valid?"
Think of Pramiti as DNS for AI agents. DNS translates human-readable names to machine-readable addresses. Pramiti translates human business concepts to machine-executable queries and validates agent actions against business rules before execution.
Two Products, One Platform
Pramiti delivers two products from a single codebase, each targeting a different buyer:
| Product | What It Does | Buyer |
|---|---|---|
| Epistom | NLQ-to-SQL semantic intelligence, knowledge model, MCP server | VP Data / Data Engineering |
| Aegis | Deterministic action validation, policy enforcement, attestation | CISO / Chief Risk Officer |
A third capability, Flight Recorder, provides cryptographically signed audit trails for all agent actions.
The Two-Plane Model
Pramiti operates across two planes over a shared semantic substrate:
-
Read Plane (Epistom) — Serves governed context. Agents ask questions like "What does revenue mean?" and get verified definitions, validated SQL, and confidence scores. When the platform cannot answer, it says "I don't know" rather than guessing.
-
Write Plane (Aegis) — Validates proposed actions. Before an agent can update a CRM record, send a Slack message, or modify a database, Aegis evaluates the action against business rules and returns a deterministic verdict: ALLOW, DENY, REWRITE, or ESCALATE.
Design Principles
- AI agents are the primary consumer. Humans curate the knowledge model. Agents query and act through it.
- Validate before execute. Every query and every action passes through a validation gate.
- Assume breach. The LLM will be compromised (prompt injection may never be fully patched). Enforcement lives outside the model, is deterministic, and operates at the action layer.
- Model-agnostic. Claude is the default LLM, but the platform supports OpenAI, AWS Bedrock, and Ollama via configuration.
- Never auto-modify the knowledge model. Drift detection and feedback produce suggestions. Humans always approve changes.
- Governance as byproduct. Nobody decides to "do governance." It happens because they used the product.
Who Is This For?
- Data teams who want AI agents to understand their business semantics, not just column names
- Security teams who need deterministic control over what AI agents can do
- Platform teams deploying MCP servers and need a trust layer between agents and enterprise systems
- Compliance teams who need regulator-grade attestation for EU AI Act and SOC 2 requirements