Deploy failed 03:10 · report ready 03:15

Production breaks fast.Beacon answers faster.

The moment something fails, Beacon starts investigating. Root cause in minutes, with evidence behind every claim.

See how it works ↓

Evidence-cited · budget-capped · every claim verified

03:12:04 AM · Beacon collects

Somewhere in 50,000 lines is one sentence that matters.

pydantic.ValidationError: 1 validation error for Settings · DB_URL: field required

03:12:06 AM · Beacon investigates

Three hypotheses. Real tool calls. One verdict.

Beacon doesn't summarize your logs. It argues with them. Every hypothesis is confirmed or refuted with evidence it fetched itself, and every claim in the report cites the line that proves it.

incident 629f64bd · meetpilot-apitriage

pydantic.ValidationError: 1 validation error for Settings · DB_URL: field required

api/config.py: DB_URL missing from the deploy environment

prior 70%

postgres: connection pool exhaustion during rollout

prior 20%

reject

deploy pipeline: stale image cache on node-3

prior 10%

reject

7 tool calls · 18,342 tokens · 4s wall clock · budget respected

03:12:08 AM · Beacon reports

You wake up to the answer, not the alert.

Every claim in the report cites the line that proves it. A deterministic gate flags any citation it can't verify against the evidence Beacon actually gathered. The model writes the prose; code guarantees the evidence.

beacon.local/incidents/629f64bd

Root cause

High confidence

DB_URL was dropped from the deploy environment, so the API fails settings validation at boot and the rollout never goes healthy.

Evidence

app.log:2093pydantic.ValidationError: DB_URL field requiredverified

api/config.py:14Settings.DB_URL has no default, required at bootverified

commit 8d1f42cdeploy.yaml: DB_URL removed from the env blockverified

Ruled out

Connection-pool exhaustion: the pool never opened, refused before the first query.

Stale image cache on node-3: the running image digest matches the built artifact.

Suggested next step

Restore DB_URL to the deploy environment and re-run deploy #4127.

citation gate · 0/ 3 citations verified

How it runs

One failed deploy. A cited answer about five minutes later.

03:12:04 AM

01 collect

The failing run reports itself

app.log

The Action grabs the trailing log window, recent commits, and run metadata the moment a deploy concludes in failure.

03:14:38 AM

02 investigate

Hypotheses meet evidence

verdicts

The agent ranks competing root causes, then verifies each one with real tool calls under a hard token and tool-call budget.

03:17:11 AM

03 report

The answer finds you

report.md

An evidence-cited report lands in the dashboard, your inbox, and as a comment on the failing pull request.

under the hood · four agents, one relay

Meet the Beacon agents

Every incident is a relay: each agent does one job, writes its result to shared state, and hands off. No agent talks to another; the pipeline decides what runs.

the pipeline · four agents, one compiled graph

One graph.

Four agents that never call each other. Each writes to shared state, and the graph routes. Click any stage to see its wiring.

STARTEND

Agent 1 · CollectorPure Python, no LLM · compression is the skill

The next failed deploy triages itself.

Sign in, point Beacon at a repo, and the next red deploy arrives with hypotheses tested, citations verified, and a report waiting.

View on GitHub →