Skip to content
Sign inGet started
Observe. Experiment. Improve.

Understand the run.
Prove the next step.

Observe your agent, experiment with improvements, and prove what works. Tracing, versioned datasets, experiments, and evaluation in one focused workspace.

Create your workspace See how it works ↓

OpenTelemetry + OpenInference. Your own private workspace.

ILLUSTRATIVE TRACE
Trace inspector / Support agent

A clear view of every decision.

Success
AGENT Resolve customer question 1.24 s
TOOL Search knowledge base 184 ms
LLM Generate grounded answer 832 ms
CHECK Validate response 12 ms
From execution to evidence.Inputs · Outputs · Hierarchy
✓ OpenTelemetry + OpenInference tracing✓ Dataset-backed experiments✓ Evidence-backed evaluation
One workspace. The complete feedback loop.

From “it ran” to
“we know it works.”

Connect the dots between what your agent does and the quality your users experience.

Follow every run.

Inspect model calls, tools, inputs, outputs, errors, and parent-child relationships. Find the part of a run that actually matters.

Trace inspection & search

Experiment with confidence.

Turn traces into versioned datasets. Test a prompt, model, or agent change and compare baseline and candidate results.

Experiments & versioned benchmarks

Prove what works.

Run your agent on test questions and judge its real answers by meaning. Basic AI judging is Free when configured and explicitly approved.

Evaluation & evidence
Designed to fit your agent

Keep your stack.
Get the bigger picture.

1

Create your account.

Verify your email and open a workspace that belongs to you.

2

Add questions. Connect your agent.

On Free, enter test questions and optional trusted references, then configure an authorized HTTP agent. No actual answers or latency to paste. Premium adds AI-written questions through an embedded Stripe subscription.

3

Run it. Judge the meaning.

Set judging criteria and approve execution. Tracework captures real agent answers and latency, then a basic AI judge evaluates meaning, correctness against supplied facts and relevance. Different wording can pass. Configuration, consent and provider charges apply.

NODE.JS + LANGCHAIN / SOURCE HELPERconst telemetry = initializeOpenInference({
  project: "my-agent",
  endpoint: "https://your-tracework/api/v1/traces",
  apiKey: process.env.TRACEWORK_API_KEY,
});

try {
  await agent.invoke(question);
} finally {
  await telemetry.shutdown();
}
Experiments are core, not an afterthought

Test the change.
Not your intuition.

An evaluation scores a result. An experiment tests a specific configuration against a fixed dataset so you can see whether the next version is actually better.

  1. Pin the benchmark.

    Use the same dataset version, evaluator, thresholds, and judge criteria for both runs.

  2. Compare baseline and candidate.

    Inspect per-example improvements and regressions alongside latency, tokens, and recorded cost.

  3. Review before adopting.

    Follow the underlying traces. Keep important regressions visible and make the final decision yourself.

See Free and Premium experiments
Illustrative experiment setup · not a live benchmark

Can a cheaper model keep quality?

Baseline

Current configuration

Existing prompt and model

Candidate

Proposed configuration

One deliberate model change

Same dataset versionSame evaluation criteriaTrace-linked evidence

Compare the trade-offs, not just the average score.

  • Quality
  • Regressions
  • Latency
  • Tokens
  • Recorded cost

Live model and agent experiments require operator configuration and explicit approval. Provider usage charges may apply.

Plans for evaluation and hosting.

Free to start.
Premium to go deeper.

Start with tracing, experiments, and evaluation. Our Premium direction adds automation, optimization, and collaboration.

Free

$0

Observe. Experiment. Establish your baseline.

Understand your agent, test a change, and inspect the evidence yourself.

  • OpenTelemetry + OpenInference tracing
  • Versioned datasets and human feedback
  • Manual experiments and basic comparisons
  • Deterministic quality and retrieval checks
  • Automatic agent runs and basic AI judging
  • Prompt versions and recorded usage metrics
  • Private workspace and agent API keys
Start with Free

Premium

Available with Stripe

Stripe subscription

Automate the improvement loop.

Subscribe in-app for AI test generation, with deeper experiments, AI-assisted improvement, and team workflows on the roadmap.

  • Everything in Free
  • Multi-configuration experiments and repeated trials
  • Advanced AI judges and calibration
  • AI failure analysis and test generation
  • Prompt optimization and CI quality gates
  • Continuous quality monitoring and cost controls
  • Team review workflows and governance
Upgrade to Premium

Hosted agent

Monthly host fee

Keep a saved agent available.

Pay monthly, for as long as you want the agent hosted. Scratch tests still expire.

  • Saved agent stays up while the fee is paid
  • No separate hosting time limit
  • Stops new runs when the subscription ends
  • The saved agent remains after you cancel
Pay the host fee

AI-generated test cases require a Premium subscription synced directly from Stripe. Other Premium capabilities include roadmap work and are not newly gated. AI calls still require configuration, dollar budgets and consent; provider charges apply.

Compare Free and Premium

Observe. Experiment.
Prove what works.

Your next improvement starts with evidence you can inspect.

Get started with Tracework