Mastra

Sazabi gave us the production visibility we needed for launch without building out our usual observability setup. Now it’s a safety net we trust every time we deploy.

About

Mastra is an open-source TypeScript framework that helps developers build, run, and observe AI agents without leaving the JavaScript ecosystem. Since launching from Y Combinator in 2025, the company has raised $35 million, including a $22 million Series A led by Spark Capital, and has helped teams at Salesforce, MongoDB, and Replit build agent-powered products and internal workflows.

Challenge

Real-Time Alerts Flagged Problems but Required Manual Diagnosis

With a conference launch approaching, Mastra’s engineers were racing to finish Trace Intelligence, a new feature that groups agent traces to help customers understand patterns in their agents’ behavior. Although Mastra was already collecting logs in Google Cloud, the team had yet to build its usual backend monitoring around the feature.

Abhi Aiyer, Co-Founder and CTO at Mastra, needed his lean platform engineering team to support the launch without losing hours to investigation. “I want them to ship code, I want them to move fast,” he shares. He already had real-time alerts across Mastra’s established systems, but those notifications marked the beginning of an investigation, not a diagnosis.

That left engineers responsible for connecting the evidence across tools. A Slack alert sent an engineer into Google Cloud to review errors, pull together the relevant logs, and then hand that evidence to an AI tool for analysis. Reaching a diagnosis could take 45 minutes. As platform usage grew each week, those investigations pulled engineers away from building new capabilities. Mastra’s alerts covered what the team already expected to go wrong. But for the new feature, the team was still “flying blind,” Abhi explains. He needed production problems to be surfaced and explained, not another tool to check.

He turned to Sazabi, which works from Mastra’s existing logs without requiring the usual monitoring setup first.

It was a 45-minute journey just to get a root cause. Sazabi takes on that investigative work, so we can focus on fixing the problem rather than piecing together what happened.

Solution

Sazabi Delivers Production Alerts With Root-Cause Analysis

Onboarding started with a permission grant, connecting Sazabi to Mastra’s Google Cloud logs. Sazabi’s agent mapped the infrastructure and assigned findings to the affected systems, with no miscategorization. Errors surfaced within seconds. “All it took was a couple of clicks, and then we could see all the things that were going wrong,” Abhi says.

Sazabi searched and correlated the logs, ranking findings by severity level. Engineers started with critical failures, such as database connectivity issues and Pub/Sub infrastructure receiving no events. Medium-level findings, like API requests returning 500 errors and unhandled JavaScript rejections, ranked lower, giving engineers a clear order for remediation. With the diagnosis already in hand, Abhi’s team can focus on validating each finding.

Alerts arrive in Slack with findings attached. Sazabi’s notification delivery rules route alerts about specific services to the channels used by the teams responsible for them. Senior engineers spend less effort on routine instrumentation, while Mastra keeps targeted checks for selected systems.

One deployment shows how that works in practice. When an engineer deployed a change that introduced a production issue, Sazabi flagged the error and linked it to the exact change with the context needed to roll back and correct it. Abhi reports deployment-related alerts typically arrive less than 15 minutes, so the team treats silence as its own signal. While engineers still test their own changes after a release, a quiet channel “gives them confidence that the deploy was good,” he notes.

When Sazabi does flag a problem, the response is wired in. Mastra connected Sazabi to incident.io so critical issues trigger paging. Every finding enters the Linear backlog, with high and critical alerts receiving immediate attention. A designated engineer rotates weekly to triage the backlog and manage less urgent work. Mastra uses its own internal agent workflow to act on Sazabi’s alert and root-cause context, moving findings toward a fix and leaving engineers time for high-priority work.

With Sazabi, we get the alert and the root cause at the same time. We can act on what’s wrong without going through three different systems to figure out why.

Results

150 Real Production Issues Identified in Two Weeks

With Sazabi, Mastra catches production failures early, freeing engineering capacity to build new features and capabilities. Sazabi helped Abhi’s team find and correct issues in Trace Intelligence, while engineers validate the findings and prioritize follow-up through Mastra’s existing response workflow.

• 45-minute investigations replaced with immediate root-cause analysis

• 150 real production issues identified in two weeks

• <5 minutes from production setup to Sazabi detecting issues

Trace Intelligence is one step toward Mastra’s broader goal of automating agent improvement, from reviewing traces to running experiments. Sazabi helps Abhi and his team respond to production issues with greater operational confidence as they continue shipping.

As usage grows, Sazabi gives us the extra help to operate the platform better and fix problems quickly for our customers.