Plurai

Plurai: simulation and post-training for AI agents

Plurai Simulation wires a digital twin of your agent into a mocked world of services, documents and synthetic users, plays the scenarios most likely to break it in an isolated sandbox, and returns a ranked fix plan with exact code locations.

How it works

• Install the plugin in Claude Code, GitHub Copilot CLI or OpenAI Codex (Cursor coming soon) and run /simulate on your agent's folder.
• Scan: Plurai maps tools, sub-agents, skills and knowledge bases and generates a harness (simulate.py) without touching your source.
• Suite: scenarios and evals spread across your agent's capabilities by weight, including hard personas and edge cases.
• Run: every scenario plays three times in a resettable sandbox; writes are mocked unless you opt in.
• Fix and verify: each gap is root-caused to files and lines, fixed, and re-run to confirm.

Works with

LangGraph, Claude Agent SDK, Google ADK, CrewAI, OpenAI Agents SDK, LlamaIndex, Strands, LangChain, Mastra, Vercel AI SDK, Salesforce Agentforce and n8n; traces from LangSmith, Langfuse, Arize and Braintrust; Python SDK for headless runs.

Post-training

Small, task-specific models for evals and guardrails (PII leaks, grounding, intent, tool-call validation) served as low-latency endpoints. The company claims about 44x lower cost than Claude Haiku-5.5.

Pricing

• Free: $0, one-time 12,300 credits, 2 agent setups, 3 runs
• Base: $60/month, 28,710 credits, 20 runs
• Business: $210/month, 65,570 credits, 60 runs

Founded by Dr. Ilan Kadar (CEO) and Dr. Elad Levi (CTO); $13M seed led by Team8.