Laminar

Laminar is an open-source observability and debugging platform for AI agents, built around OpenTelemetry. It traces LLM calls, tool use, custom functions, and sub-agents, showing activity in readable transcripts. Users can describe outcomes or failures in plain language with Signals, which extract structured events across traces. For debugging, the platform can replay a run from cache up to the point under review, using cached results for earlier calls. Evaluations against datasets can run locally or in CI to help catch regressions. Laminar also provides realtime traces, full-text search, raw SQL access through its UI, CLI, or MCP, and custom dashboards. It offers managed cloud and self-hosting; self-hosted deployments can run on AWS or GCP with Helm charts. The Free plan is free and includes 1 GB of data, seven-day retention, one project, and one seat. Starter is $30 a month. Privacy Mode varies by plan, and the stated PII redaction is best-effort.

Laminar combines agent tracing, debugging, and evaluation tools with managed and self-hosted deployment choices. Review plan limits and privacy settings, particularly the best-effort scope of PII redaction, before choosing it.

Laminar suits teams building AI agents that need to trace model and tool activity, investigate runs, or evaluate changes against datasets. It offers both managed cloud and self-hosted deployment options.

Laminar is an open-source observability and debugging platform for AI agents, suited to teams tracing work across models, tools, and sub-agents. Its strongest case is the combination of readable traces, cached replay, and dataset evaluations, with cloud and self-hosted deployment choices. Plan quotas and privacy settings deserve attention before teams send it sensitive data.

Overview

Laminar follows LLM calls, tool use, custom functions, and sub-agents, then presents those runs as readable transcripts. That breadth is useful when a failure emerges from an agent’s sequence of actions rather than from a single model response. Signals let teams describe outcomes and failures in plain language and turn them into structured events across traces.

Realtime traces, full-text search, custom dashboards, and raw SQL access through the UI, CLI, or MCP give teams several ways to inspect activity. The platform is OpenTelemetry-native and documents integrations for OpenAI, Anthropic, Gemini, Cohere, LiteLLM, Vercel AI SDK, LangChain, Pydantic AI, Claude Code, Codex, Browser Use, Stagehand, and Playwright. That range makes it a plausible fit for varied agent stacks, though a specific integration should be checked against a team’s implementation needs.

Key features

Replay and evaluation

The debugger replays a run from cache up to the point under investigation, serving earlier calls from cache. This can make iterative debugging less wasteful when the fault appears late in a multi-step run. Dataset evaluations can run locally or in CI, giving teams a way to catch regressions during development as well as in an automated pipeline.

Data access and privacy

Raw SQL access and custom dashboards provide room for teams to inspect trace data beyond standard views. Laminar states that it is HIPAA and SOC 2 Type II compliant and offers server-side PII redaction on eligible plans. Privacy Mode is off by default on Free and Starter, on for Pro and higher, and on and unchangeable for accounts covered by a signed DPA. When enabled, Customer Data is not used to train or improve Laminar’s machine-learning models. PII redaction is best-effort and covers model input and output content, so it should not be treated as the sole protection for sensitive data.

Pricing

Laminar is freemium, with paid plans starting at $30/mo. The free plan is free and includes 1 GB of data with no overage, a one-time $5 Signals credit, seven-day retention, one project, one seat, and community support. It is a useful way to try the core workflow, but the single project and seat constrain team use.

Starter costs $30 a month and includes 3 GB of data, then charges $2 / GB; it also has a one-time $5 Signals credit, 30-day retention, unlimited projects and seats, and email support. It removes the free plan’s collaboration caps, but data overage and relatively short retention matter for growing workloads.

Pro costs $150 a month and includes 10 GB of data, then charges $1.50 / GB. It provides a one-time $5 Signals credit, six-month retention, unlimited projects and seats, and Slack support. The longer retention and lower overage rate suit teams that need a deeper history, though the monthly step up from Starter is substantial.

Enterprise has custom pricing and custom limits, with on-premise deployment, unlimited projects and seats, and dedicated support. It is the option for organizations that need those deployment and support terms; the plan’s usage limits and cost depend on the custom arrangement.

Platforms

Laminar is available through an API, on the web, and as a self-hosted deployment. Self-hosted installations can run on AWS or GCP using Helm charts, giving infrastructure teams a route to operate it in their own environment. Managed cloud is also offered for teams that prefer not to self-host.

Who it's for

Laminar is best suited to developers and teams building AI agents whose behavior spans model calls, tools, and sub-agents. Cached replay and local or CI evaluations are especially relevant to teams diagnosing multi-step failures and guarding against regressions. The managed and self-hosted options also serve teams with different infrastructure preferences. A solo user can start on Free, while teams should account for the one-seat cap there and the paid plans’ data quotas.

Pros and cons

• Pros: Traces cover LLM calls, tools, custom functions, and sub-agents, making the product relevant to complex agent runs.
• Pros: Cached replay and evaluations against datasets support debugging and regression checks across local and CI workflows.
• Pros: Managed cloud and self-hosting, including AWS or GCP deployment with Helm charts, give teams deployment flexibility.
• Cons: Free is limited to one project, one seat, 1 GB of data, and seven-day retention, which narrows its usefulness for ongoing team work.
• Cons: Starter and Pro charge for data beyond their included quotas, so larger trace volumes can raise the bill.
• Cons: Privacy Mode is off by default on Free and Starter, and PII redaction is best-effort, requiring teams to manage sensitive data carefully.

Alternatives

For an open-source framework with a free plan, consider Evidently AI; its listed platforms include API, web, and self-hosted use. Langfuse is another freemium option, with a free Hobby plan that includes 50k units per month, 30 days of data access, and two users. Teams comparing free agent observability choices can also consider Maxim AI, Memoturn, Opik, Qualdo-MQX, W&B Weave, and Arize AX. Browse the Model Monitoring Software, LLM Observability Tools, and AI Agent Observability Tools lists for more options.

Verdict

Choose Laminar if your team needs to trace agent behavior across tools and sub-agents, then investigate runs with replay and dataset evaluations. Its open-source foundation and managed or self-hosted deployment make it adaptable to different teams. Look elsewhere if the free plan’s project and seat caps are too restrictive, if expected data volume makes overages a concern, or if your privacy requirements depend on PII redaction as a guaranteed safeguard.