Laminar

• CollectionObservability & Tracing
• Classificationagent-native
• LicenseApache 2.0 (self-host) · Y Combinator-backed · $3M seed (March 2026)
• MCPUnavailable
• Agent SkillUnavailable
• InterfacesCLI · SDK · API

• Website ↗
• Docs ↗
• Official repo ↗
• Full dossier on GitHub ↗

What it does

Laminar is an agent-native observability platform. Where conventional LLM observability tools (Langfuse, LangSmith) optimize for prompt → completion pairs, Laminar’s data model is agent-trace-first: thousands of spans, non-deterministic control flow, nested causality, and session continuity are first-class.

Five capabilities define the product:

1. One-line tracing for popular AI frameworks and SDKs.
2. True agent debugger — local agent runs visible in the browser; rerun at step N with previous context preserved; tune system prompts and reflect changes instantly.
3. Browser-agent session replay — automatic capture and sync with traces (Browser Use, Stagehand, Playwright, Kernel, Browserbase).
4. Signals — describe an event in natural language; Laminar extracts it from past and future traces (e.g., “agent failure due to logic error”), with a defined output schema and email/Slack reports.
5. SQL over traces — query the entire trace corpus with SQL; feed Evals datasets straight from a query.

Self-host is fully Apache 2.0; the platform is written in Rust and aimed at terabyte-scale trace data.

Agent-facing interfaces

• Catalog protocols: CLI, SDK, API
• MCP: ⚠️ Not yet published as a first-party MCP server. Laminar’s data plane is OpenTelemetry-compatible; tracing is integrated via the SDK.
• Agent Skill: ⚠️ Not yet published as an official skill.
• Onboarding type: cli

How agents onboard

# Self-host:
git clone https://github.com/lmnr-ai/lmnr
cd lmnr && docker compose up -d

# Or hosted: sign up at laminar.sh

# Two lines to integrate:
from lmnr import Laminar
Laminar.initialize()

See the full onboarding section in the dossier.

Protocol surface