The trace is the alert.
No alert leaves without the failing request attached. A page with no trace is a bug in our product, not a feature.
00 / 05 · About
Next Quarter, Nobody Hunts Through Logs At 3 A.M.
Every alert arrives with its trace. The slow hop is already named. That is the product. Everything else serves it.
- Protocols REST · GraphQL
- Unit of work One request
- Alert payload Trace attached
- Probe regions 25
- Plans Spindle · Labyrinth · Minotaur
01 / 05 · The problem
The Page Said “Slow.” It Did Not Say Where.
A request crosses seven services. Your alert knows one.
LOG-01 / the alert
A threshold trips. A phone buzzes.
Latency crossed a line. The page names the endpoint and stops there. Nothing about which service held the request.
LOG-02 / the hunt
Five dashboards. Four tabs of logs.
Someone greps by timestamp. Someone else guesses at a deploy. Clocks disagree between services. Twenty minutes go by.
LOG-03 / the fix
It was one row lock.
The fix took four minutes. Finding it took the rest of the night. We built Wayfinder’s Thread to delete the middle step.
02 / 05 · Principles
Four Rules We Ship By.
They decide what gets built. They decide what gets cut.
Follow the request, not the host.
Servers come and go. A request has a start, a path and an end. We model that.
Fewer pages. Better pages.
Mute windows, SLO burn rates and deploy context keep the noise down. A page should mean someone acts.
Pricing you can read in one pass.
Three plans, billed annually. The tier sets endpoints and retention. The core tracing stays the same.
FIG.01 — support.queue · help center
03 / 05 · How it’s built
Engineers Answer The Help Queue.
Support is staffed by the people who write the agent. A question about span naming reaches someone who knows why spans are named that way.
The agent is small and tuned to stay out of your request path. Probes run from 25 regions on a one-minute loop. Traces stay on your plan’s retention clock, then they’re gone.
- Agent: one install line per service
- Retention: 7 days to 365 days by plan
- Help center: email, answered by engineers
04 / 05 · Where it’s going
The Route Ahead, In Plain Rows.
What we are working on. Plans change when you tell us something we missed.
RD-01 / now
Deeper GraphQL spans
Per-field resolver timing inside a single query, so one slow resolver stops hiding in a fast gateway.
BuildingRD-02 / next
Deploy-aware SLO alerts
Burn-rate alerts that name the deploy before the budget runs dry, not after.
NextRD-03 / next
More agent runtimes
Wider language and runtime coverage, so a trace never breaks at a service we can’t see into.
NextRD-04 / later
Shareable trace links
Send a read-only trace to a vendor whose service is the slow hop.
Later05 / 05 · Your move
Pick A Plan. Follow The Thread.
Three plans. One thread. Tonight’s first slow request shows its whole route.
Get started