Proxy Health

IDE proxy traffic and box resources. Refreshes every 20s.
Loading...

Traffic 5-minute buckets from proxy_requests

n/a
Requests
n/a
People active
n/a
Errors (status not ok or client_closed)
n/a
client_closed
n/a
Cost (USD)
n/a
Cost per request

Requests per bucket by source

People active per bucket

Errors and client_closed per bucket

Latency latency_ms includes CLI queue wait

p50 (solid) and p90 (dashed) latency by source, seconds

Latency by path_mode

Warm CLI plan telemetry baseline. warm_spawn/warm_continue appear once the sidecar is wired in; until then every row is cache, failover, cold_fresh, or cold_resume.
path_modeRequestsp50 latencyp90 latencyp50 TTFTp90 TTFTCost/request
Loading...

Cache and resume efficiency

n/a
Cache-read share
n/a
Continuity hit rate
n/a
Cache read tokens
n/a
Cache write tokens
n/a
Uncached input tokens
n/a
Cost per request

Input tokens per bucket

Cache-read share = cache read / (cache read + cache write + uncached input).

Cache-read share and continuity hit rate, %

Continuity hit rate counts only rows that carry a continuity verdict.

Cost per request, USD

Per engineer, sorted by cost

PersonRequestsContinuity hitCache-read shareCostCost/requestp50 latency
Loading...

Efficiency token-efficiency plan, from anthropic_usage; 14-day trend always shown regardless of the selected window

n/a
Cost per request
n/a
Cache-write share
n/a
1-hour share of writes
n/a
Background share of calls
n/a
Background cost
n/a
Downgraded opus/fable calls

Cost per request, 14-day trend

Cache-write share, 14-day trend, %

Per engineer

PersonRequestsCostCost/requestWrite shareBackground callsBackground costDowngraded callsDowngraded cost
Loading...

Background shapes

routed_small_model = calls the router actually downgraded to a cheaper model, not just labeled.
ShapeRouteCallsCostCost/callRouted to small model
Loading...

Distrust last 2h only

Responses matching the distrust pattern

Regex over response_body. Capped to the last 2 hours regardless of the selected window.
n/a

By source

SourceMatchesRows with bodyRows
Loading...

Pattern

Box resources live read on the serving host, not stored

n/a
Load average 1 / 5 / 15
n/a
RAM used
n/a
RAM available
n/a
Live claude CLI processes
n/a
claude CLI total RSS
n/a
Redis used memory

systemd units (cgroup v2)

CPU% is percent of one core, from the cpu.stat usage_usec delta between samples.
UnitCPU%memory.current
Loading...

Warm CLI sessions integrations/warm_cli/* sidecar - off by default (MAESTRO_PROXY_WARM_CLI); n/a until the unit is installed

n/a
Idle
n/a
Busy
n/a
Awaiting tool
n/a
Total warm RSS
n/a
Warm hit share warm_continue / (warm_continue+warm_spawn+cold_resume)
n/a
Spawns since sidecar boot

Evictions by reason since sidecar boot

ReasonCount
Loading...

Fallbacks by reason since sidecar boot

ReasonCount
Loading...

Live warm sessions

EngineerAccountStatePIDAgeIdleTurnsRSS MB
Loading...

Endpoint health

proxy_requests status by source

SourceStatusRows
Loading...

count_tokens and 5xx by path

count_tokens status countsn/a
5xx by path (excluding /health)n/a