Mastra launches hosted Observability platform with 100k free events a month

Tracing and monitoring for agents in production

Mastra Observability is aimed at agents that are already live. After a team connects a project, Mastra automatically traces agent runs, workflows and tools. The setup uses an Observability configuration in the Mastra project with a platform exporter. According to the company, traces are batched and sent to the platform asynchronously in the background. A SensitiveDataFilter can redact sensitive fields before traces are exported.

Users can search and filter traces, then inspect model and tool calls to see where time and tokens were spent and what they cost. A thread view places each run in the context of the full conversation. Mastra says this lets non-technical teammates start with the messages and drill into the underlying steps without first decoding a raw span tree.

For monitoring, teams can run online scorers on a sample of their traces to track correctness automatically. A dashboard shows trends over time next to performance and cost metrics.

Eval loop, AI-assisted debugging and CI checks

The platform also supports an eval loop. Teams can review conversations, annotate traces and curate datasets, then run an experiment against a dataset and compare agent versions on quality, cost and latency. In the example shown in the post, the author tested whether a downgrade to nano could cut costs without hurting performance, and found that it could not.

Mastra points to growing volumes of trace data as agent harnesses take on bigger jobs and longer runs. Its query traces CLI gives a coding agent a way to investigate slow, expensive or failed runs and help fix the underlying issues. A further feature, Trace Intelligence, is in private beta and available on request. It surfaces recurring failure patterns by grouping traces by goals, behavior, sentiment and outcomes, so teams can prioritize fixes and pick examples for evals.

Evals can also run in code. Gates and Verdicts turn behavior checks and scorer thresholds into a pass/fail verdict that a test suite can assert on, which helps catch regressions in CI/CD before shipping. Mastra says teams can optionally wrap this in a GitHub Action so every pull request runs the evals and shows the verdict as a check.

Pricing and what it means for teams

Mastra is offering 100k observability events free each month, with 15 days of retention and no credit card required. Additional usage costs $10 per 100k events. The Teams plan costs $250/month and includes:

• 1m events per month
• six months of retention
• multiple teams
• SSO
• SOC 2 documentation

Extra usage on the Teams plan is $8 per 100k events.

For teams already building on the Mastra framework, the release puts tracing, scoring, datasets and CI checks in one hosted product tied to their existing project. Mastra also targets teams on other tools: a trace import guide is available, and the company offers help bringing traces and datasets over to Mastra.