multi-router.aidocs

Observability

Traffic

The Traffic page: views, time ranges, filters, columns, stats, charts and CSV export.

Traffic is the dashboard page that lists every request the organization's routers recorded, newest first. It opens from Traffic in the top navigation, showing the last 24 hours. Views, time ranges and filters narrow the list, a strip of stats and charts summarizes the requests in view, and any request opens in place to show its full detail.

Views

Engineering · org_7Tq2Wd9KxP

Overview · Traffic

CustomizeSave view
Time:
1h24h7d30d
CustomFilters
Search any id (request/correlation/session)…
Export
Requests
128,403
Success rate
99.4%
5xx errors
142
Tokens
84.2M
Latency · p95
1,240ms
Latency · p99
3,180ms
Cost
$1,284.50
Customers
312
Requests · Count / time128,403
Outcome · 5xx / time142
Latency · p95 / time1,240ms
Cost · Total / time$1,284.50
Requests
Page 125 / page
Columns
DateTimeRequest IDRouterAliasModelVendorIn/OutCostLatencyStatus
2026-09-3014:32:11req_Lw4Tn8Vb2ksupport-bot—anthropic/claude-sonnet-4.5anthropic2,310/512$0.011.94s200
2026-09-3014:32:09req_9Pc1RmYe4srt_9Hk3Ws7PzQ—anthropic/claude-haiku-4.5bedrock860/210$0.0021640ms200
2026-09-3014:32:07req_8Fq2LmXo1crt_9Hk3Ws7PzQ—anthropic/claude-haiku-4.5bedrock1,240/380$0.0035808ms200
2026-09-3014:32:04req_3Zr6KdWq8nclaude-code-eng—anthropic/claude-sonnet-4.5vertex5,120/1,034$0.033.21s200
2026-09-3014:32:02req_7Hb5NsTq3wdraft-classifier—openai/gpt-5-miniopenai640/42$0.0002410ms200
2026-09-3014:32:00req_2Kd8VxRe6mclaude-code-eng—anthropic/claude-opus-5.5anthropic12,480/2,206$0.128.42s200
2026-09-3014:31:58req_6Ym2HcPa5tsupport-bot—anthropic/claude-sonnet-4.5anthropic0/0$0212ms429
2026-09-3014:31:55req_5Wt9GpLc1zdraft-classifier—openai/gpt-5.4openai0/0$012.7s502
2026-09-3014:31:52req_4Qn3BvYk7rrt_9Hk3Ws7PzQ—anthropic/claude-haiku-4.5bedrock910/264$0.0024690ms200
2026-09-3014:31:49req_8Jv1MzDs2psupport-bot—anthropic/claude-sonnet-4.5anthropic3,045/688$0.022.48s200
The Traffic page with the Overview view: the views rail, time range and filters, stats, charts and the Requests table. Select to enlarge.

A view is a saved combination of filters, time range, stats and charts. The views rail on the left lists the built-in views and the person's own saved views:

ViewStats and charts for
OverviewRequest volume, success, latency and cost across all requests.
LatencyTotal latency, time to first token and the time spent at the provider.
ErrorsError rates and error kinds.
CostTotal cost and its parts.
By modelRequests, latency and cost broken down by model.
CustomersUsage broken down by the caller's customer id.
FeaturesRequests and cost broken down by feature label, and cost by model.

A built-in view sets the stats and charts and applies to every request in the time range and filters. Save view stores the current filters, time range, stats and charts as a saved view, which can be renamed or deleted. Changing a saved view marks it Modified, and saving updates it. A saved view keeps the length of its time range, such as the last 7 days, and applies it from the moment it is opened.

Time range

The time range sets which requests the page shows. The Time control offers 1h, 24h, 7d and 30d, and Custom offers these ranges:

RangeKind
Last 15 minutesRelative
Last 1 hourRelative
Last 6 hoursRelative
Last 24 hoursRelative
Last 7 daysRelative
Last 30 daysRelative
TodayCalendar
YesterdayCalendar
This weekCalendar
Last weekCalendar
This monthCalendar
Last monthCalendar

The page shows up to 30 days of requests. It loads when the range, filters or page size change, and a new choice of range brings in the latest requests.

Filters

Filters narrow the requests to those that match every filter set. The Filters menu offers:

FilterMatches
ModelThe model that served the request.
RouterThe router the request entered through, or the member router that served it, by id.
ProviderThe provider that served the request.
StatusAny, 2xx, 4xx or 5xx.
StreamingAny, Yes or No.
Error kindThe error kind exactly as recorded, such as rate_limit_error or timeout. See Request Detail.
CustomerThe caller's customer id, such as tenant_acme.
SessionThe session id, exactly as sent in x-mr-session-id.
CorrelationThe correlation id, exactly as sent in x-mr-correlation-id.
FeatureThe feature label sent in X-MR-Feature-Id. Exact matches the whole label, and Prefix matches its start, so support/ matches support/triage and support/reply.
Machine issuerThe identity provider of a Machine Identity workload.
Machine subjectThe workload identity, such as a service account.
Machine bindingThe machine binding the workload matched.
Router typeAny, llm, or fallback.

The search box finds a request by any of its ids: the request id, a correlation id or a session id, matched exactly.

Requests table

The Requests table lists one request per row, 50 per page by default, with 25 and 100 also available, and the page buttons move to the first, previous, next and last page. Selecting a request id opens the request in place. See Request Detail. The Columns button chooses which columns appear, and the choice is remembered in the browser:

ColumnShown by defaultShows
DateYesThe date of the request.
TimeYesThe time of the request.
Request IDYes, alwaysThe request id, which opens the request.
RouterYesThe router that served the request.
AliasYesThe alias the request used, when it used one.
ModelYesThe model that served the request.
VendorYesThe provider that served the request.
In/OutYesInput and output tokens.
CostYesThe cost of the request.
LatencyYesTotal latency.
StatusYesThe status code.
TypeNoThe router type.
VariantNoThe offering variant, such as standard or global.
RegionNoThe region that served the request.
Wire idNoThe provider's own model id.
CustomerNoThe caller's customer id.
FeatureNoThe feature label from X-MR-Feature-Id.
Identity typeNoThe kind of Machine Identity workload, when one made the request.
Machine issuerNoThe identity provider of the workload.
Machine subjectNoThe workload identity.
Machine bindingNoThe machine binding the workload matched.
AttemptsNoHow many attempts the request took.

Stats and charts

The stat strip and chart strip above the table summarize the requests in view. Customize chooses up to 8 stats and 4 charts, in any order, and a chart can be broken down by a dimension, such as latency by model. Latency metrics offer p50, p90, p95, p99, maximum and average; throughput offers average and p95; attempts offer average, maximum and p95. The metrics are:

MetricGroup
RequestsRequests
Success %Outcome
Error rateOutcome
4xxOutcome
5xxOutcome
Rate-limitedOutcome
RetryableOutcome
Recovered %Outcome
Substituted %Outcome
Upstream error rateOutcome
Upstream 5xxOutcome
Total latencyLatency
Time to first tokenLatency
Upstream latencyLatency
Time spent in Multi-RouterLatency
Total tokensTokens
Input tokensTokens
Output tokensTokens
Cache read tokensTokens
Cache create tokensTokens
Cache hit %Tokens
Tokens per requestTokens
Throughput, output tokens per secondThroughput
Total costCost
Cost per requestCost
Cost per 1,000 tokensCost
Input, output, cache read and cache write costCost
AttemptsAttempts
Multi-attempt %Attempts
Distinct models, providers, routers, customers, features, sessions and error kindsDistinct
Busiest model, provider, router, customer, feature and error kindBusiest

A chart can break down by one of these dimensions:

Dimension
Model
Provider
Router
Customer
Feature
Error kind
Session
Machine issuer
Machine subject
Machine binding
Router type

Export

Export downloads the requests that match the current time range and filters as a CSV file named for the range, newest first, with up to 100,000 rows. A larger export ends with the line # truncated at 100000 rows, and a narrower time range or more filters fetch the rest. The file has these columns:

CSV columnstext
timestamp, request_id, alias_name, router_id, leaf_router_id, model_served, provider,
status_code, prompt_tokens, completion_tokens, cost_usd_micros, latency_total_ms, attempts,
customer_user, router_type, classifier_type, classifier_version, calibration_version,
difficulty_score, raw_complexity, task_type, classify_latency_ms, short_input_floored,
model_group, ladder_size, rung_index, first_choice_model, token_estimate, has_code,
turn_count, gateway_platform, variant, region, model_wire_id, identity_kind, issuer_id,
subject, binding_id, integration_outcome, integration_blocked_by, integration_blocked_by_name,
latency_integrations_us, integration_runs, stream_block_offset, stream_chars_sent,
parent_request_id, account_id, person_id, api_type, streaming, user_agent, stop_reason,
provider_key_id, provider_request_id, provider_response_id, reasoning_tokens, inference_geo,
trace_id, client_tool_call_count, attempt_provider, attempt_model_slug, attempt_variant,
attempt_region, attempt_model_wire_id, attempt_provider_key_id, attempt_upstream_api_type,
attempt_start_time, attempt_duration_ms, attempt_first_byte_ms, attempt_status_code,
attempt_error_kind, attempt_provider_request_id, attempt_provider_response_id, feature_id,
rule_id, rule_name

cost_usd_micros is the cost in millionths of a US dollar, so 1500 is $0.0015. router_id is the router the request entered through, and leaf_router_id is the member router that served it. feature_id is the feature label sent in X-MR-Feature-Id. The columns from classifier_type to turn_count are kept for older records and are empty for current requests.

OverviewRequest Detail
Was this helpful?