Laminar Agent Observability

LMNR AI, Inc., which does business as Laminar, develops an open-source observability platform for AI agents, with a San Francisco address in its terms of service. It addresses the difficulty of finding unexpected failures in long-running agent runs. Its blog announced a $3 million raise in March 2026, and the homepage lists Y Combinator in the footer.

The platform records LLM calls, tool calls and sub-agents in a readable trace view, then analyses every run through Signals that surface failure modes and group them into clusters. A CLI and MCP server give coding agents SQL access to traces, and error clusters convert into eval datasets. Browser Use is quoted as a customer.

• Trace view: captures LLM calls, tool calls, sub-agents, costs and tokens in a readable transcript.
• Signals: analyses every agent run to surface unanticipated failure modes.
• Signal clusters: groups similar failures into patterns with their distribution.
• Slack alerts: notifies teams about new failures and lets them query traces.
• Debugger: CLI and MCP server give coding agents SQL access to investigate traces and verify fixes.
• Evals: turns error clusters into datasets and reruns them after changes.

A Vietnamese startup building a browser-automation agent can cluster recurring failures and turn them into regression tests. A fintech support team can query agent traces through Slack. A software house can self-host the open-source version.

Failure clusters surface problems that teams did not anticipate, and they convert into eval datasets to catch regressions. The vendor says its storage format is 20x cheaper by keeping only unique content. A free plan costs $0 with 1 GB of data, and paid plans start at $30 per month. SOC 2 Type II and HIPAA compliance are listed.

Agents that run for long periods produce very large traces, so teams need automated failure analysis instead of reading logs by hand. Vietnamese software firms building AI agents for browsing, support or back-office tasks face the same debugging burden.

Legal entity LMNR AI, Inc. (DBA Laminar), 2261 Market Street, San Francisco; terms governed by Delaware law. $3M raise announced March 2026; flow-1 model launched October 2026. Apache 2.0 open source; SOC 2 Type II and HIPAA listed.