v0.1.0: Initial Public Release

AgentTrust v0.1.0

Challenge-response quality verification for AI agents and MCP servers.

Highlights

• 3-level evaluation pipeline: Manifest → Functional → Domain Expert
• 6-axis scoring: accuracy, safety, reliability, process quality, latency, schema quality
• Consensus judging: 2-3 LLM judges with early-stop agreement (saves 50-66% LLM calls)
• 7-provider fallback chain: Cerebras → Groq → OpenRouter → Gemini → Mistral → DeepSeek → OpenAI
• 5 adversarial probe types: prompt injection, PII leakage, hallucination, overflow, system prompt extraction
• Battle Arena: head-to-head evaluation with OpenSkill (Bayesian ELO) ratings
• IRT Adaptive Testing: Rasch 1PL calibration, reduces eval cost by 50-90%
• W3C Verifiable Credentials (AQVC format) with Ed25519 signatures
• Google A2A v0.3 native support
• MCP Server with 4 tools for IDE integration (Claude, Cursor, Windsurf)
• 552 tests, all mocked

Quick Start

# Docker
cp .env.example .env && docker compose up -d

# Local
pip install -r requirements.txt
python -m uvicorn src.main:app --port 8002

# MCP package
pip install mcp-agenttrust

See README for full documentation.