Laminar
github.com/lmnr-ai/lmnr- Category
- AI Agents
- Rank
- No. 74Tools index
- Pricing
- Open Source
- Platform
- web · cli
- Type
- TOOL
- Builder
- lmnr-ai
- GitHub
- 3.2k stars
- Latest release
- v0.2.1
- Added
- Jul 4, 2026
About
Laminar is an open-source observability platform built specifically for AI agents. It lets you trace every agent run, set up plain-English alerts for failure conditions, and run evaluations to catch regressions — all in one place. It integrates with Vercel AI SDK, LangChain, OpenAI, Anthropic, and more with just a line of code.
What it does
Laminar turns agent activity into searchable operational data. It accepts OpenTelemetry traces, shows live runs, supports full-text and SQL exploration, builds dashboards, and converts annotated production data into datasets for later evaluation work.
Why it's ranked here
The breadth is credible because the repository exposes real ingestion, storage, querying, alerting, evaluation, and privacy machinery. It suits teams that want one operational loop from production traces to debugging and regression checks, while retaining a managed or self-hosted deployment choice.
What's good
The backend accepts HTTP and gRPC traffic, separates ingest-only credentials from broader project access, and caps public SQL queries to protect ClickHouse. Redis failures can fall back to in-memory cache and pub/sub. The separate PII redactor bounds request size and token count, batches concurrent inference, and reports an unhealthy worker as unavailable.
Tradeoffs
Self-hosting is more than installing an SDK. The backend depends on PostgreSQL and can involve ClickHouse, Redis, RabbitMQ, and an optional redaction service. AI-assisted features require separate provider credentials. Anonymous telemetry is enabled for self-hosted deployments unless disabled. Sharing PostgreSQL with another Drizzle-managed service may require manual migration work.
How to use it well
Use it for TypeScript or Python agent systems where production traces should feed debugging, annotation, datasets, and CI evaluation. Start with managed hosting or the lightweight Compose stack, then add full infrastructure when load justifies it. It does not provide an LLM; its AI-assisted features need Gemini, OpenAI-compatible, or Bedrock access.
Technical notes+
app-server/src/main.rs assembles Actix HTTP and tonic gRPC servers, OpenTelemetry trace and log ingestion, PostgreSQL, cache and pub/sub selection, queue workers, ClickHouse paths, and feature-gated signals. app-server/src/auth/mod.rs distinguishes ingest-only API keys from broader project keys. app-server/src/sql/mod.rs validates queries before routing them to cloud ClickHouse or a hybrid data plane, with separate treatment for internal and public traffic. app-server/src/cache/mod.rs and app-server/src/mq/mod.rs define in-memory or Redis caching and Tokio or RabbitMQ queue abstractions. pii-redactor/src/main.rs implements a bounded, batched tonic service with configurable model location, chunking, overlap, token limits, and request limits.
Observed
- Implementation language
- The provided backend and PII redaction service are written in Rust.
- SDK packaging
- Client SDKs install from npm as @lmnr-ai/lmnr and from PyPI as lmnr with optional instrumentation extras.
- Interfaces
- The documented surfaces include SDKs, a CLI, MCP access, HTTP endpoints, and gRPC ingestion.
- Deployment
- Laminar offers a managed platform and self-hosting through Docker Compose, including a fuller production Compose configuration.
- Storage and infrastructure
- The shown backend uses PostgreSQL and ClickHouse, with Redis or in-memory caching and RabbitMQ or Tokio queue implementations.
- Telemetry
- Self-hosted deployments collect anonymous usage telemetry by default and expose an environment setting to disable it.
Read from README.md, app-server/src/main.rs, pii-redactor/src/main.rs, app-server/src/ch/mod.rs, app-server/src/db/mod.rs, app-server/src/mq/mod.rs, app-server/src/api/mod.rs, app-server/src/env/mod.rs, app-server/src/llm/mod.rs, app-server/src/sql/mod.rs, app-server/src/auth/mod.rs, app-server/src/logs/mod.rs, app-server/src/cache/mod.rs, app-server/src/names/mod.rs, app-server/src/utils/mod.rs.
What it can do
Trace every run of an AI agent end-to-end
AI agent code instrumented with a one-line SDK integration → Detailed execution traces showing each step, input, output, and timing of the agent run
Alert on AI agent failure conditions written in plain English
A natural-language description of a failure condition (signal) → Real-time alerts triggered when the agent behavior matches the defined failure condition
Run evaluations to detect regressions in AI agent behavior
Agent runs and defined evaluation criteria → Evaluation results highlighting behavioral regressions across agent versions
Integrate observability into AI frameworks with a single line of code
Existing code using Vercel AI SDK, LangChain, OpenAI, Anthropic, or similar → Fully instrumented agent with automatic tracing enabled
Self-host the observability platform
Docker Compose configuration → A privately hosted Laminar instance running on the user's own infrastructure
Access agent observability data via CLI or MCP for coding agents
CLI commands or MCP protocol requests → Queried trace data, signals, or evaluation results returned programmatically
Tags
Media
Comments (0)
No comments yet
Indexed by a proprietary survey. Corrections welcome.