Ship reliable agents
Laminar is an open-source agent observability platform.
It automatically catches agent failures and helps you fix them.
See clearly what your agent is doing
Laminar automatically captures , , sub-agents, costs, and tokens, and shows it in a readable transcript view.
Learn more about the trace viewDiscover failures without defining them
Laminar Signals analyze every agent run to surface failure modes you didn't anticipate.
Learn more about SignalsSimilar failures are clustered into distinct patterns to give you a high level overview.
Understand agent failures at scale
Signal clusters show the full distribution of your agent's failures and behaviors. Understand whether a newly reported issue has occurred before.
Learn more about Signal ClustersThe agent ran three times for the same question, carried on past a 404 from without retrying, then answered from a snippet without linking the page it read.
See clearly what your agent is doing
Laminar automatically captures LLM calls, tool calls, sub-agents, costs, and tokens, and shows it in a readable transcript view.
Learn more about the trace viewThe agent ran three times for the same question, carried on past a 404 from without retrying, then answered from a snippet without linking the page it read.
Discover failures without defining them
Laminar Signals analyze every agent run to surface failure modes you didn't anticipate.
Learn more about SignalsThe agent ran three times for the same question, carried on past a 404 from without retrying, then answered from a snippet without linking the page it read.
Similar failures are clustered into distinct patterns to give you a high level overview.
The agent ran web_search three times for the same question, carried on past a 404 from fetch_page without retrying, then ai.llm answered from a snippet without linking the page it read.
Understand agent failures at scale
Signal clusters show the full distribution of your agent's failures and behaviors. Understand whether a newly reported issue has occurred before.
Learn more about Signal ClustersBe notified and
investigate in Slack.
Laminar notifies you about new failures and clusters in Slack.
Mention Laminar to ask anything about your traces.
#laminar-alerts
Claude, fix my agent with Laminar
The Laminar CLI and MCP give your coding agent everything it
needs to debug your agent. Query traces, evals, signals with SQL,
replay from any checkpoint, verify the fix.
Did the new release break anything?
Every error cluster you fix can automatically be turned into an eval dataset.
Run evals after a change to catch regressions and iterate with confidence.
Two lines to integrate with your stack
Comprehensive platform for agent development.
Automatic agent input extraction
Laminar extracts the agent's task from every run. Use it to easily create e2e eval datasets from production traces.
Add rate limiting to the /v1/traces ingest route. Reject over-quota projects with a 429.
I should add a token bucket keyed by project id, then check it in the handler before the batch is parsed.
{ "path": "src/routes/traces.rs" }
Custom dashboards
Build dashboards to track statistics from traces and signals with custom SQL queries.
Subagents per run
4.34
New traces
12.4k
Top span names
Total tokens
48.2M
Full SQL access
Query all platform data with raw SQL. Have your coding agent query data with MCP or CLI.
SELECT model,
count(*) AS runs, avg(duration) AS p50_ms,
sum(total_tokens) AS tokens, sum(total_cost) AS cost
FROM spans WHERE span_type = 'LLM'
GROUP BY model ORDER BY cost DESC
UI for fast data annotation
Build labeled datasets from traces, dataset rows, or SQL results for evals and fine-tuning.
Screen recording for browser agents
Capture your agent's browser session alongside the trace.
flights.example.com/offer/48
Full-text search
Full-text search across every span input, output, and attribute.
output
The request hit a timeout at the booking gateway after 30s.
Retried once and saw the same timeout, then fell back to the
cached fare quote from the previous search.
attributes
error.type = timeout
error.message = upstream timeout after 30000ms
http.status_code = 504
http.route = /v1/offers/search
retry.count = 1
gen_ai.request.model = gpt-5.1
“We run millions of agent sessions in our cloud,
and when something goes wrong,
Laminar’s trace view is the first place we look”
Magnus Müller, CEO
20x cheaper storage
Laminar stores only the unique content in agent runs, leading to faster ingestion and 20x cheaper storage.
Read moreCompetition
Laminar
Open source from day zero
GitHub stars
Apache 2.0 license
Set up with Docker
Deploy on AWS or GCP with Helm charts
~ $
Ready for Enterprise
Ship reliable agents