Skip to main content
AgentMark is a fully local-first agent engineering platform that helps you build, test, and monitor AI applications with confidence.

​
What You Get

Prompt Management - Write prompts in readable Markdown + JSX
  • Declarative syntax that shows exactly what your LLM sees
  • Reusable components and templating
  • Version control-friendly format
  • Hot-reload during development
Testing & Evaluation - Ensure quality before production
  • Datasets for systematic testing
  • Custom evaluation functions
  • CLI and SDK for running experiments
  • CI/CD integration
Observability - Monitor and debug in production
  • Distributed tracing with OpenTelemetry
  • Session tracking for multi-step workflows
  • Token usage and cost monitoring
  • Integration with AgentMark platform
Local-First - Everything runs on your machine
  • No vendor lock-in
  • Full control over your data
  • Works offline
  • Optional cloud sync with AgentMark platform

​
Core Features

Readable Prompts: 
<System>You are a helpful math tutor.</System>
<User>What's {props.num1} + {props.num2}?</User>
Systematic Testing: 
npm run experiment math-tutor.prompt.mdx
Production Monitoring: 
telemetry: {
  isEnabled: true,
  metadata: { userId, sessionId }
}

​
Why AgentMark?

For Development:
  • Decouple prompts from application code
  • Test prompts systematically with datasets
  • Iterate quickly with hot-reload
  • Works with any AI SDK (Vercel AI, Claude Agent SDK, Mastra, Pydantic AI)
For Production:
  • Monitor performance and costs
  • Track token usage
  • Debug user issues
  • Analyze usage patterns

​
Supported Languages

LanguageAdapters
TypeScriptAI SDK (Vercel), Claude Agent SDK, Mastra
PythonPydantic AI, Claude Agent SDK

​
Have Questions?

We’re here to help! Choose the best way to reach us: