AI NewsInfrastructureAnnouncement
Limnion launched Siglake, an Apache 2.0 telemetry lake for AI agents
Limnion launched on Y Combinator with Siglake, an Apache 2.0 telemetry lake. It stores logs and traces as open-format files on a team's own object storage. An AI agent can query the data as SQL. The founder co-founded InfluxData.
Image: Limnion
Why it mattersA team's observability bill often climbs faster than its traffic. Open-format storage on a team's own bucket lets any SQL agent read the data. The team keeps control of the files.
A team watching its Datadog or Splunk bill knows the shape. Logs and traces live in a vendor's store. An AI agent that could answer "what broke last night" has to go through the vendor's API. Limnion launched on Y Combinator this week with a different shape. Its core product, Siglake, is an Apache 2.0 telemetry lake. Logs and traces live as Parquet and Iceberg files on the team's own object storage. Any SQL tool can read them.
The founder is Todd Persen. He co-founded InfluxData, which built the InfluxDB time-series database. The YC batch is not stated on the launch page. Limnion lists its investors as array.vc, Foundation Capital and Y Combinator. The company says Siglake stores telemetry in open formats (Parquet, Iceberg, Arrow) and makes it queryable with SQL by default.
What is new
Three parts matter. First, Siglake takes data in through OpenTelemetry, which is the standard protocol most observability tools already speak. Second, the store is a team's own S3 or GCS bucket, in Parquet files on an Iceberg table. Third, the query engine is DataFusion SQL. An agent that already writes SQL for a data warehouse can read production logs in the same way.
The Apache 2.0 licence is the point. A team can self-host Siglake without a licence fee. The company's hosted service will be the business model. The pricing is not published yet.
Why agents matter here
Siglake is designed so an AI agent can read the data without a human in the middle. An on-call engineer asks "which deploy caused this spike". An agent writes the SQL and runs it. The agent sees the same table a human sees. The schema is OTel's, which is public.
Limnion cites Gartner, Datadog and its own view. Gartner, by the company's reading, projects the observability market at about $14 billion by 2028. Datadog's own guidance, which the launch page cites, is about $4.5 billion in revenue this year. Limnion also says 44 percent of Gartner clients spend more than $1 million a year on observability. Those are external figures, attributed. Limnion predicts that "90% of human-centric observability work disappears within two years". That last line is the company's own forecast, not a measurement.
Who it is for
Two teams. One, a team running an AI product in production that wants an agent to triage incidents on its own. Two, a team on a large Datadog or Splunk bill that already pipes data through OpenTelemetry. The second case is the easier migration. A team that already sends OTel data can swap the destination.
The honest limit is that a self-hosted observability stack is work. Siglake stores the data. A team still needs dashboards, alerts and the on-call rotation. Vendor products include those as part of the price. A team moving to Siglake inherits the pager duty of running the lake.
Source
- Primary source: Limnion on Y Combinator.
- Company site: Limnion.
- Repository: siglake/siglake on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
