Monitor
Know how your agents are doing at a glance.
Support agent Live
Interactions38,472Last 7 days
Success rate96.4%↑ 2.8% this week
See every agent.
Improve every answer.
Backed by
Glass connects traces, failures, and evaluations so your team can understand what happened and make your AI agents better.
Monitor
A clear view of how your agents are performing.
LiveG
An answer goes wrong. Know why, and what to do next.
Production evidenceOne connected workflow
Every answer has a story.
Glass shows what happened, what keeps happening, and whether your next change makes it better.
Trace
Follow an answer through every model and tool step.
Interaction trace1.28 s
“Can I change my delivery address?”
Look up order policy184 ms
Before-dispatch condition missedReview
Generate answer412 ms
Signals
Turn repeated mistakes into one clear issue.
Open signalNeeds review
Incorrect delivery policy
Agent omits the “before dispatch” condition.
12interactions
3similar questions
2hsince first seen
Evals
Make a real failure the test for your next release.
Delivery address changesCase from production
Expected answer
“You can change it before your order is dispatched.”
1 case added to the regression suite
Daily digests
Get the wins and warning signs in one update.
GlassDaily digest · 9:05 AMImproved
p95 latency down 30.2%
Needs attention
89 tool retry loops detected
Regressed
Task failures rose this week
The feedback from our customers has been very positive. The insights have helped us better understand what happens in our chats and identify significant issues.
agent.pyPython
# Add Glass to your existing agent
import os
from glass import init, trace
init(api_key=os.environ["GLASSAI_API_KEY"])
@trace()
def generate_response(prompt):
return agent.run(prompt)Frequently asked questions
GlassAI works with any LLM-powered agent — whether you're using OpenAI, Anthropic, Mistral, or an open-source model. We support multi-step pipelines, tool-calling agents, RAG workflows, and custom orchestration frameworks.
Under 2 minutes. Install our SDK, add one line of instrumentation to your agent, and you'll start seeing traces immediately — no infrastructure changes required.
We only store the trace data you explicitly send us — inputs, outputs, and metadata from your agent runs. You control what gets logged and can mask or redact sensitive fields before they leave your environment.
Security is a top priority for us. We are currently in the process of obtaining SOC 2 Type II and ISO 27001 certifications. In the meantime, all data is encrypted in transit (TLS 1.2+) and at rest (AES-256), and we follow industry best practices for access control and vulnerability management. Enterprise customers can request our security documentation and sign a DPA.
Yes. We offer a self-hosted option for teams with strict data residency or on-premise requirements. Reach out to us to discuss your setup.
The free plan includes full observability for up to 10,000 traces/month. Paid plans unlock higher volume, anomaly detection alerts, eval pipelines, team collaboration, and priority support. See the pricing page for details.
Give every answer a way to get better.
Bring traces, signals, and evals into one clear view.