Laminar

About Laminar

Laminar is an open-source observability platform built specifically for agents rather than for LLM calls, and the distinction shows in its core primitive. Instead of asking you to write assertions, it lets you describe a failure in plain English — "agent is stuck in a loop" — as a Signal; Laminar then reads every agent run, evaluates it against that description and pings Slack when the pattern occurs. That inverts the usual observability workflow, where you only catch the failures you thought to instrument. Once alerted, the debugging path is designed to be short: the run is rendered as a readable transcript and timeline surfacing inputs, LLM reasoning, tool calls and sub-agents, and you can ask questions of the run directly and get answers that cite specific steps. Around that sit the rest of an evaluation stack — OTLP trace ingestion, full-text trace search, custom dashboards, a SQL editor, datasets, labelling queues, browser session recording and MCP access. Billing is metered on two axes that reflect how the product works: gigabytes of trace data, and dollars of Signals evaluation spend, with published input and output token rates for the overage. The free tier is a real one at 1 GB and $5 in Signals with no overage, and paid tiers offer SOC 2 Type II, HIPAA and server-side PII removal.

Key Features

Laminar Pros & Cons

Tags