Agentic Veritas

Sister in the Agentra ecosystem | .veritas format | 20 Capabilities | 10 MCP Tools | 30+ CLI Commands

Why AgenticVeritas

Every AI agent guesses what you mean. You say "build an API" and the agent picks a framework, a database, an auth strategy, and a deployment target -- without asking. When the result does not match your intent, you blame the model. But the model was never given a precise specification of what you wanted.

The current fixes do not work. System prompts give context -- but they cannot resolve ambiguity. Few-shot examples show patterns -- but they cannot detect when the user's intent is genuinely uncertain. Chain-of-thought reasoning explains steps -- but it cannot flag the assumptions it made along the way.

Current AI: Guesses intent and presents results with false confidence. AgenticVeritas: Compiles intent into precise specifications, flags uncertainty, resolves ambiguity through clarification, and verifies claims against evidence.

Quickstart

cargo install agentic-veritas-cli
veritas --help

Problems Solved (Read This First)

• Problem: agents interpret ambiguous prompts by guessing, producing results that miss the user's intent. Solved: intent compilation parses natural language into structured domain, entity, constraint, and operation specifications -- ambiguity is detected before execution.
• Problem: agents present all answers with equal confidence, even when they are uncertain. Solved: uncertainty detection scores every claim with calibrated confidence, and flags statements that lack supporting evidence.
• Problem: agents cannot distinguish verified facts from inferred assumptions. Solved: truth verification separates claims into verified, supported, unsupported, and contradicted categories with evidence chains.
• Problem: agents cannot reason about cause and effect in user requirements. Solved: causal reasoning traces dependency chains between requirements, detecting when one decision forces or prevents another.
• Problem: there is no way to know what questions the agent should have asked but did not. Solved: ambiguity detection identifies gaps, presents clarifying questions, and uses defaults transparently when the user does not respond.

# Compile what you mean, verify what is true -- three commands
veritas compile "Build a REST API for task management with team collaboration"
veritas ambiguity detect <intent-id>
veritas verify <intent-id>

How It Works

Architecture Overview

+-------------------------------------------------------------+
|                     YOUR AI AGENT                           |
|           (Claude, Cursor, Windsurf, Cody)                  |
+----------------------------+--------------------------------+
                             |
                  +----------v----------+
                  |      MCP LAYER      |
                  |   10 Tools + stdio  |
                  +----------+----------+
                             |
+----------------------------v--------------------------------+
|                   VERITAS ENGINE                             |
+-----------+-----------+------------+-----------+------------+
| Intent    | Ambiguity | Uncertainty| Causal    | Truth      |
| Compiler  | Detector  | Scorer     | Reasoner  | Verifier   |
+-----------+-----------+------------+-----------+------------+

20 Capabilities

Install

git clone https://github.com/agentralabs/agentic-veritas.git
cd agentic-veritas
cargo install --path crates/agentic-veritas-cli

curl -fsSL https://agentralabs.tech/install/veritas | bash

Standalone guarantee: AgenticVeritas operates fully standalone. No other sister, external service, or orchestrator is required.

MCP Server

{
  "mcpServers": {
    "agentic-veritas": {
      "command": "agentic-veritas-mcp",
      "args": ["serve"]
    }
  }
}

License

MIT -- see LICENSE.