Find out which bug ships next — before your users do.
crxes reads the logs you already collect and runs four AI agents over them in sequence: parse, detect patterns, trace root cause, predict. What comes back isn't another alert — it's a short list of the failures forming right now, with the confidence and the ETA attached.
Free tier · no credit card · read-only credentials, encrypted at rest
- Log Parser1.2s
- Pattern Detector3.4s
- Root Cause Analyzer5.1s
- Bug Predictor4.6s
Checkout writes will start timing out
payments-api connection pool exhaustion — pool_size=20 against a checkout worker count raised to 64 in the 14:02 deploy. Wait-time p99 has climbed 8x in 40 minutes.
Recommended action
Raise SQLALCHEMY_POOL_SIZE to 64 on payments-api, or roll back the worker-count bump in deploy 8f21c4.
How it works
Three steps from credentials to a forecast.
Connect a source
Walk the wizard, paste read-only credentials, hit Test connection. We validate the key against the provider and show you the sample log count before anything is saved.
Pull logs
Pull now fetches on demand. Or set a schedule — every 15 minutes through every 4 hours — and Celery Beat keeps pulling without you.
Read the forecast
The pipeline runs and streams its progress live. Three to five predictions land, ranked by severity, each with the root cause behind it.
The pipeline
Four agents, each handing off to the next.
One model asked to “find bugs in these logs” gives you a summary. Splitting the job into four specialists — each with a narrow brief and the previous one's output as its input — is what turns raw lines into a claim specific enough to act on.
Log Parser
Normalizes every provider's payload into one shape — timestamp, level, service, message — and throws away the noise.
Pattern Detector
Clusters the normalized lines into recurring signatures and flags the ones whose rate is bending upward.
Root Cause Analyzer
Traces each rising cluster back through the surrounding lines to the change or dependency that explains it.
Bug Predictor
Turns causes into forecasts: what breaks, how badly, how confident, how soon, and what to do first.
Progress streams over SSE — you watch each agent land, not a spinner.
Sources
Point it at the logs you already have.
No agent to install, no shipper to reconfigure. Give crxes read-only credentials and it pulls on your schedule. Credentials are encrypted at rest and never leave your account.
Datadog
Logs API, scoped by query and index
CloudWatch
Log groups via an IAM read-only key
Sentry
Issues and events per project
Google Cloud
SoonCloud Logging filters
Grafana Loki
SoonLogQL against your Loki instance
Webhook
SoonPush anything to a signed endpoint
Once it's running
Built to run unattended, and to be checked afterward.
Scheduled pulls
Pick 15 min, 30 min, hourly, or 4-hourly per source. Celery Beat owns the cadence; you own the dial.
Auto-analyze
Fire the full pipeline after every scheduled pull, so predictions are waiting for you instead of the other way round.
Live agent progress
Server-sent events push each agent's start and finish as it happens — no spinner, no polling, no guessing where it is.
Browsable history
Every analysis and prediction is stored. Filter by source, severity, or date and reopen any run with its full agent output.
Accuracy feedback
Mark a prediction accurate or inaccurate with a note. Your calls become the scorecard for how much to trust the next one.
Source health
The dashboard surfaces failing credentials, stale pulls, and paused sources before a silent gap turns into a blind spot.
Pricing
Start free. Upgrade when the schedule matters.
Free
Enough to find out whether the forecasts hold up.
$0
- ·1 connected source
- ·25 analyses per month
- ·Manual pulls
- ·7 days of history
Pro
For sources you need watched around the clock.
$29 / month
- ·Unlimited sources
- ·Unlimited analyses
- ·Scheduled pulls from every 15 minutes
- ·Auto-analyze after each pull
- ·Full history and accuracy tracking
Your logs already know what's about to break.
Connect a source and get your first forecast in about two minutes.