We finally fixed Observability!
Legacy stacks were built for humans reading dashboards. Berserk is built for agents asking questions — keep every event at a fraction of the cost. Fast, schemaless, OTEL-native, petabyte-scale, and self-hosted.
Let your agents do the digging, so your engineers can do the building.
Built by veterans from
Why Berserk?
Headless by Design
Dashboards show you what you already know. Agents find what you don't. Berserk pairs with Claude Code, MCP, and the bzrk CLI to investigate incidents in seconds.
Built for AIOps
Built-in CLI and MCP server. Query in KQL — the language LLMs already speak. Streaming, recency-first results, so agents act on signal without waiting for the full scan.
Read more →Zero-Config Ingest
Point your OpenTelemetry collectors at Berserk and ship. Schemaless ingest means no migrations when your agent stack changes weekly.
Keep Everything, Forever
10× more telemetry for the same spend. Retain every event, log, and trace indefinitely — not just what you remembered to instrument — and answer questions you didn't know to ask. Your data, your cloud, your object store.
Trace-Joins
Filter, join, and aggregate across billions of traces — application requests and agent runs — at petabyte scale. Analysis other telemetry engines can't express.
Read more →Compliance Posture
- GDPR-compatible
- ISO 27001-compatible
- NIS2-compatible
Start with a question, not a dashboard
Dashboards only answer what someone already thought to ask. Berserk lets you — or your agent — just ask. Three ways in:
Claude Code
Describe the symptom in plain English. Claude Code writes the KQL, reads the traces, and follows the thread across services until it finds the cause.
Set it up →MCP Server
Berserk speaks the Model Context Protocol natively, so any MCP-aware agent or IDE can query your telemetry directly — no glue code, no exports.
bzrk CLI
One binary, KQL on the command line. Wire it into CI, runbooks, and cron — or pipe results straight into the next tool in the chain.
“My inner nerd is just really happy right now — bzrk + Claude Code is so lovely. We had a prod crash-loop and Claude decided to use the bzrk:incident-triage skill by itself to figure out what was up, and it did. +1”
Integrations
Observability needed a rewrite
Software is producing more telemetry than ever. Security events, error logs, application traces—and now AI prompts and model reasoning logs. With every deploy, the volume grows. Together they tell the story of our systems, but the plot gets lost in the noise and scale.
AI agents make the problem even harder. Their prompts and reasoning generate large, text-heavy outputs that capture the decisions driving our businesses. Prompts and model reasoning don't fit traditional telemetry schemas, yet they carry the same operational signal.
What we need isn't an auxiliary system for AI logs. We need a unified system that can correlate logs, metrics, traces, and AI output.
And we need a system that doesn't require a human at a dashboard. Dashboards answer questions someone already thought to ask. Modern incidents — and agentic ones especially — demand questions nobody anticipated. Berserk gives agents the data, the query language, and the latency to ask them.
We built Berserk for telemetry in the AI era. It is schemaless, fast, and designed to handle large text-heavy logs alongside traditional telemetry—while remaining exceptionally affordable, even at petabyte scale.
Ready to get started?
Run the full stack locally with Docker Compose in under five minutes.