Skip to main content

Agent Lens

Trust your agents. Verify with evidence — conversationally, on MLflow.

Agent Lens is a conversational qualification layer that drives the upstream official MLflow MCP so platform engineers can evaluate, qualify, and govern AI agents in natural language.

tip

Works with any agent framework that sends traces to MLflow — LangGraph, Google ADK, LangChain, CrewAI, OpenAI Agents SDK, and more.

The Problem

You have 50 AI agents in production. Some you built, most you didn't. MLflow gives you traces and scorers — but no one is systematically grading them, no one is blocking bad deployments, and compliance has no audit trail.

The Solution

Ask Agent Lens in plain English:

You: "Evaluate financial-advisor-agent using the RAG profile"
You: "Show me traces with errors in the last 24 hours"
You: "Can this agent be deployed to production?"

Agent Lens calls MLflow MCP tools, runs evaluations, and returns structured verdicts.

Architecture

MLflow = data plane (traces, scorers, models) Agent Lens = decision plane (verdicts, governance, fleet management)

Your agents can be built with any framework — LangGraph, Google ADK, LangChain, CrewAI, OpenAI Agents SDK, or custom. If they send traces to MLflow, Agent Lens can qualify them.

Quick Start

git clone https://github.com/rrbanda/agent-lens.git && cd agent-lens

# Local development with integration tests
make dev-setup # Create venv, install deps
make mlflow-start # Start local MLflow server
make seed-data # Populate with test traces
make test-integration # Run integration tests against real MCP

# Deploy to OpenShift
make deploy-all # Build image + deploy Sandbox
make status # Verify MCP + pod health

Who It's For

PersonaTheir QuestionWhat Agent Lens Gives Them
Agent Platform Engineer"Can I qualify this agent?"Qualification verdicts, fleet observatory, trace forensics
Agent Developer"Will my agent pass quality gates?"CI/CD gate API, eval-in-pipeline, regression tracking
Chief AI Officer"Is this investment paying off?"Fleet-wide quality scores, executive summaries
CISO / Security Lead"Is the security boundary robust?"Governance audit trail, policy violation tracking
Domain Expert / SME"Did the agent do the right thing?"Trace annotation, expectation authoring