Laminar¶

Development line: project:laminar · thread laminar-development
Last event: - · 0 dated since - · Researched: 2026-09-05 · confidence: high

What it is¶

Laminar is an OpenTelemetry-native platform for agent builders, and an alternative to Langfuse or Helicone rather than an LLM gateway.

• Tracing: trace LLM and tool calls.
• Browser sessions: inspect interactive runs.
• Evaluations: run evaluation datasets.
• Signals: detect recurring failures.
• SQL: query execution data directly.
• Breakpoints: replay a failed run.

Self-hosting requires a multi-service stack. Signals requires a Google Generative AI API key.

Development line¶

• We have not written up the dated timeline yet; what we know stands in the sections below.

What changed¶

• 2026-03-16 — Laminar announced a $3 million seed round and positioned the product around debugging and monitoring long-running agents: trace transcript views, browser-session replay, step-level debugger, Signals, SQL analysis, and evaluation datasets.
• 2026-03-18 — The reported funding was $3 million, led by Atlantic.vc with participation from Y Combinator, AAL.vc, Ben Sigelman, and Ant Wilson; the product scope was agent observability rather than a model release.
• 2026-07-09 — Release v0.2.1 added a refreshed Clusters UI, configurable frontend base path, API-key expiration metadata, CLI connection in onboarding, and Signals alerting changes.

How to use this¶

As of 2026-03-18, make no implementation or adoption change yet; first verify the linked Laminar announcement, repository state, and model reference.

1. Create a project and project API key, install the TypeScript SDK, and call Laminar.initialize at application startup before making model calls. — https://github.com/lmnr-ai/lmnr
2. Wrap agent entrypoints and meaningful application functions with observe so the trace contains the workflow rather than only a root span. — https://github.com/lmnr-ai/lmnr
3. For Python, install lmnr with only the instrumentations required by the application, initialize it once early in startup, and supply LMNR_PROJECT_API_KEY through the environment. — https://github.com/lmnr-ai/lmnr-python
4. Choose managed hosting for the shortest path, or self-host with Docker Compose; use the full Compose configuration for production workloads. — https://github.com/lmnr-ai/lmnr
5. Review traces and Signals, then use the built-in SQL editor to turn production observations into datasets and run evaluations against them. — https://laminar.sh/blog/2026-03-16-laminar-launch

Best practices¶

• Initialize the SDK once early in startup, rather than repeatedly per request or tool call. — https://github.com/lmnr-ai/lmnr-python
• Preserve an existing OpenTelemetry span design and point its OTLP exporter to Laminar instead of duplicating instrumentation. — https://github.com/lmnr-ai/lmnr/blob/main/frontend/assets/blog/2026-03-05-migrate-from-helicone-to-laminar.mdx
• Use the lightweight Docker Compose stack only for quick starts or light usage; use managed hosting or docker-compose-full.yml for production. — https://github.com/lmnr-ai/lmnr
• Do not treat Laminar as a replacement for gateway functions such as provider routing, response caching, or rate limiting. — https://github.com/lmnr-ai/lmnr/blob/main/frontend/assets/blog/2026-03-05-migrate-from-helicone-to-laminar.mdx

Superseded by this¶

• 2026-03-16 — The initial launch description is incomplete for deployment planning: the current project has v0.2.1 release changes, including configurable frontend base paths and API-key expiration metadata.

Still unknown¶

• The supplied Hugging Face short link could not be retrieved, so we did not use it as evidence.
• The schema omitted event_findings and new_events fields, so we incorporated their supported facts into What changed.

Sources¶

Agent brief¶

• Subject: project:laminar, thread laminar-development, 0 dated events - → -.
• Practical note: As of 2026-03-18, make no implementation or adoption change yet; first verify the linked Laminar announcement, repository state, and model reference.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.