AgentMark

AgentMark is a web platform and open-core SDK for developers and AI teams who need to build, test, and monitor AI agents in production with the same reliability loop used for traditional software.

Key capabilities include:

• Git-backed prompt, dataset, and eval management, storing prompts (MDX, Markdown + JSX) and datasets (JSONL) in the team’s own GitHub repository for auditable changes and CI workflows
• Pre-deploy evaluations in CI, so regressions can block pull requests before changes reach production
• Production observability via OpenTelemetry (OTLP), capturing traces, spans, token usage, tool calls, and metrics such as cost, latency, and error rate
• Alerts and diagnostics, firing when quality, cost, or latency crosses thresholds, then connecting incidents back to specific prompt or model changes

Pricing: Free to start (no credit card required); paid plans include Growth at $59/month and Team at $499/month. Enterprise plan is also available for custom pricing.

What makes it notable: It avoids proprietary SDK lock-in by using git for version control and standard OpenTelemetry for observability, while keeping prompts and eval assets as editable files in an existing codebase. The core SDK and prompt format are MIT-licensed on GitHub, with additional cloud features such as multi-user access, managed storage, alerting, and hosted observability.