Marker

About Marker

Marker is an evaluation platform for voice and chat agents, with deployment flexibility as its main structural bet. It runs two loops. The first improves the agent: connect the version you are changing, exercise it with real traffic and lifelike simulations, evaluate every transcript against the same set of Markers, then fix the prompt, tools, model or workflow. The second loop improves the measurement itself, which is the part most eval products skip — route the right machine evaluations to humans for review, capture a label a human will stand behind, measure judge agreement against that label on the same coordinate, and refine the judges, simulations and monitors accordingly. Both rule-based and LLM-judge markers are supported, and voice simulation is a first-class capability rather than chat evaluation with audio bolted on. The deployment story is the differentiator for regulated buyers: the same image set runs hosted by Marker, inside your own VPC where transcripts, audio and evidence never leave your account and access controls, or fully on-premises beside private systems and models, with a zero-egress air-gapped mode as a stated first-release target for self-managed installs. Moving between modes does not require changing product. Enterprise installs ship signed images, a Helm chart and an offline licence, and support bring-your-own identity provider and models. Every plan gets the full platform with no feature gates — the tiers differ only by included credits and by where the software runs.

Key Features

Marker Pros & Cons

Tags