Marker

The product is an evaluation platform for voice and chat agents that simulates conversations, scores transcripts against versioned rubrics, and calibrates automated judges with human labels to identify failures before release and in production. It is for developers, engineering teams and data teams in businesses that build and operate AI agents. It is delivered as SaaS with deployment options including hosted, customer VPC, on-premise, and air-gapped with zero egress where the data plane stays in the customer environment.

Key features

• Simulate voice and chat agent conversations
• Score transcripts with versioned rubrics
• Calibrate LLM judges with human labels
• Run regression tests across agent versions
• Ingest live transcripts with audio and timing
• Monitor tool calls and trace context
• Measure judge-human agreement rate
• Automated alerting via Slack and GitHub
• API, CLI, and MCP access
• Deploy in hosted, VPC, on-prem, air-gapped