Skip to content

Continual
learning for
agents.

Turn production failures into your next improvement. Find the issue, test a change, and carry what works into the next release.

Connect your agent See it in action
Python and TypeScript SDKs OpenTelemetry native
Follow a failure to a tested change.Interactive demo · Sample data
Acme workspace ProductionDemo

Production overview

How your agents are doing, and what needs a closer look.

Sessions24,891Across 3 agents
Success rate98.6%+0.7 pts today
Avg. latency1.24sEnd to end
Flagged sessions362 recurring issues

Session volume

Completed sessions
2k1k0
00:0006:0012:0018:0023:59

Needs attention

2 signals

Start with a signal. Follow it back to the sessions that explain it.

Monitor production. Learn from failures. Test the next change.Sample workspace
Observe. Understand. Improve. Repeat.Explore a recorded agent run
BUILT AROUND REAL AGENT BEHAVIOR

Learn from production.
Improve the next run.

Production tells you where your agents fall short. Use those sessions to understand the problem, test a fix, and check for failures you haven’t seen yet.

OBSERVE

Trace every session

See the prompts, tool calls, and responses in each session. Track latency and token costs at every step.

Explore tracing
UNDERSTAND

Catch silent regressions

Find hallucinations, tool failures, and quality drops across your agents. Group related sessions to see what went wrong.

Explore signals
IMPROVE

Test the next version

Turn real sessions into test cases. Generate edge cases, compare prompts and models, and check that your fix holds up before release.

Explore experiments
Keep the full picture.Multi-step workflows · Tool calls · Conversations · Evaluations
CONTINUAL LEARNING

Your agents should get better with use.

Build a feedback loop around real use. Review the evidence, keep the failure as a test, and compare your change before you ship it.

  1. 01
    Start with real sessions

    See where an agent struggled and what the user needed.

  2. 02
    Find the pattern

    Connect related failures to a problem you can fix.

  3. 03
    Test beyond the original issue

    Add generated edge cases and compare the change against your baseline.

  4. 04
    Bring it back to production

    Monitor the next release. Keep the tests. Repeat.

START IN YOUR TERMINAL

Add Sentrial
to your agent.

The setup wizard detects your project, guides instrumentation, and checks your connection. Or start with a few lines of Python or TypeScript.

npx --yes --package @sentrial/sdk@0.8.1 sentrial setup
Run from your agent’s project directory.Read the documentation
PythonTypeScriptOpenTelemetry
QUICKSTART
$npm install @sentrial/sdk
agent.ts
import OpenAI from 'openai';import { configure, wrapOpenAI } from '@sentrial/sdk'; configure({  apiKey: process.env.SENTRIAL_API_KEY,}); // Track calls before adding session boundaries.const client = wrapOpenAI(new OpenAI(), {  trackWithoutSession: true,});
Calls through this client are tracked automatically.
PRICINGMONTHLY PLANS

Start small.
Grow from there.

Choose a plan based on your trace volume.
Upgrade as your usage grows.

Developer

$10/ month

For side projects and testing.

Get started
  • 10,000 traces / month
  • Basic evaluations
  • 7-day data retention

Startup

FOR PRODUCTION
$450/ month

For production AI applications.

Get started
  • 1,000,000 traces / month
  • Advanced evaluations
  • Automated issue grouping
  • 30-day data retention

Enterprise

Custom

For larger teams and deployments.

Contact sales
  • Unlimited traces
  • Custom data retention
  • Dedicated support channel
  • VPC deployment options

Need a custom plan? Get in touch