HomeProductsAgenticLens

Every step,
in focus.

Trace multi-agent execution, understand token and cost usage, evaluate behavior, and turn operational evidence into better decisions. Run locally, with no mandatory hosted backend.

pip install agenticlens
  • Open source
  • Local-first evidence
  • OpenTelemetry-native
Interactive example

Follow every agent, model, and tool call.

Watch a distributed trace unfold, from delegation to retrieval and generation, with token usage and latency for the same example run.

Distributed trace · support-agentSimulated
0Tokens consumed
0.12 sRun latency
1 / 4Completed spans
  • supervisor → researchcomplete120 ms
  • research → search_docsrunningretrieving
  • research → llm.generatequeued—
  • supervisor → respondqueued—
RECENT ACTIVITY

trace: demo-7f2 · Parent span propagated to research agent; search_docs started.

Example data, simulated in your browser. No live telemetry or real actions.

BUILT FOR THE AGENT STACKLangGraphOpenAIAnthropicMCPAny Python agent
THE OPERATING LENS

From opaque runs to
release confidence.

One local toolkit connects execution evidence, quality evaluation, operational performance, and optimization decisions.

01

Trace

See every agent, handoff, tool call, retry, token, and latency event in one structured run.

See it in action →
02

Evaluate

Score outputs and execution behavior with deterministic checks or your own LLM judge.

See it in action →
03

Compare

Measure baseline and candidate runs to catch token, latency, cost, and reliability regressions.

See it in action →
04

Optimize

Turn evidence into recommendations for prompts, RAG, memory, tools, models, and handoffs.

See it in action →
05

Gate

Enforce quality, pass rate, latency, cost, and failure thresholds before production.

See it in action →
06

Export

Keep portable JSON, CSV, Markdown, Jira-ready output, and standalone HTML reports locally.

Read the docs →
INTERACTIVE PRODUCT TOUR

Trace. Evaluate. Gate.

Explore the same evidence flow used by the offline LangGraph supervisor demo included in the repository.

STRUCTURED EXECUTION

Follow the whole agent graph—not just model calls.

AgenticLens records hierarchical spans for delegation, retrieval, memory, tools, validation, retries, and final responses.

  • Agent and framework identity
  • Parent-child execution tree
  • Tokens, cost, and latency evidence
Evaluation runregional-case-total
1

SupervisorDelegate · research_agent

2

ResearchRetrieve · regional data

3

MathTool · 18 + 24 → 42

4

ReviewerFinal · grounded answer

UNIFIED EVALUATOR FRAMEWORK

Your models.
Your rubrics.
One contract.

AgenticLens does not lock evaluation to one provider. Register a trusted evaluator, reference it in a versioned suite, and route every score through the same report and release gate.

DeterministicLLM judgeSemanticSafetyRAGCustom
judge.pysuite.yaml
from agenticlens.evaluation import (
  EvaluatorRegistry, LLMJudgeEvaluator
)

registry = EvaluatorRegistry()
registry.register(
  LLMJudgeEvaluator(
    "quality_judge", call_your_model
  )
)

report = evaluate_suite(
  suite, samples, registry=registry
)
●Provider-neutral · Normalized score · Gate-ready
HOW IT WORKS

Instrument once.
Use the evidence everywhere.

AgenticLens produces portable artifacts your team can inspect, compare, share, and enforce.

01

Instrument

Wrap workflows and spans with a small Python API.

02

Observe

Capture traces, tokens, latency, cost, and behavior.

03

Evaluate

Run versioned suites with built-in and custom evaluators.

04

Ship

Compare candidates and enforce release policies in CI.

OPEN SOURCE · MIT LICENSE

Your agent data
stays yours.

No mandatory account. No hosted backend. No data egress required just to inspect a run. Start locally and integrate on your terms.

terminal
$ pip install agenticlens
$ agenticlens inspect trace.json

Run Summary
Status      succeeded
Spans       7
Tokens      86
Latency     5.2 ms

$ agenticlens gate evaluation.json
✓ Release gate passed.

Put AgenticLens to work.

Start with the documentation, explore the examples, or contribute on GitHub.