feat(copilot): server-side mothership tool/vfs/file metrics (#5071)

Author: Sg312

apps/sim/instrumentation-node.ts

@@ -83,6 +83,17 @@ function normalizeOtlpMetricsUrl(url: string): string {
   }
 }
 
+// deployment.environment in the GO value space (dev | staging | prod) without
+// any new infra env var. Every deployed Sim tier already gets
+// APPCONFIG_ENVIRONMENT = the infra env name (dev | staging | production), so we
+// reuse it and map production -> prod to match Go's `OTEL_DEPLOYMENT_ENVIRONMENT`
+// (and thus a single Grafana $env filter spans Sim + Go). Returns undefined when
+// unset (local dev) so the OTEL_/NODE_ENV fallbacks still apply.
+function deploymentEnvFromAppConfig(v: string | undefined): string | undefined {
+  if (!v) return undefined
+  return v === 'production' ? 'prod' : v
+}
+
 // Sampling ratio from env (mirrors Go's `samplerFromEnv`); fallback
 // is 100% everywhere. Retention caps cost, not sampling.
 function resolveSamplingRatio(_isLocalEndpoint: boolean): number {
@@ -270,11 +281,15 @@ async function initializeOpenTelemetry() {
       resourceFromAttributes({
         [ATTR_SERVICE_NAME]: telemetryConfig.serviceName,
         [ATTR_SERVICE_VERSION]: telemetryConfig.serviceVersion,
-        // OTEL_ → DEPLOYMENT_ENVIRONMENT → NODE_ENV; matches Go's
-        // `resourceEnvFromEnv()` so both halves tag the same value.
+        // OTEL_ → DEPLOYMENT_ENVIRONMENT → APPCONFIG_ENVIRONMENT (mapped to the
+        // Go value space) → NODE_ENV. Matches Go's `resourceEnvFromEnv()` so a
+        // single $env spans Sim + Go. APPCONFIG_ENVIRONMENT (already set on every
+        // deployed tier) is the fix that stops deployed Sim tagging everything
+        // "production" via the NODE_ENV fallback — no new infra env var needed.
         [ATTR_DEPLOYMENT_ENVIRONMENT]:
           process.env.OTEL_DEPLOYMENT_ENVIRONMENT ||
           process.env.DEPLOYMENT_ENVIRONMENT ||
+          deploymentEnvFromAppConfig(process.env.APPCONFIG_ENVIRONMENT) ||
           env.NODE_ENV ||
           'development',
         'service.namespace': 'mothership',

apps/sim/lib/copilot/chat/payload.ts

@@ -37,6 +37,7 @@ interface BuildPayloadParams {
   userTimezone?: string
   userMetadata?: {
     name?: string
+    email?: string
     timezone?: string
   }
   includeMothershipTools?: boolean
@@ -367,7 +368,8 @@ export async function buildCopilotRequestPayload(
     ...(params.workspaceContext ? { workspaceContext: params.workspaceContext } : {}),
     ...(params.userPermission ? { userPermission: params.userPermission } : {}),
     ...(params.userTimezone ? { userTimezone: params.userTimezone } : {}),
-    ...(params.userMetadata && (params.userMetadata.name || params.userMetadata.timezone)
+    ...(params.userMetadata &&
+    (params.userMetadata.name || params.userMetadata.email || params.userMetadata.timezone)
       ? { userMetadata: params.userMetadata }
       : {}),
     // Tell the copilot file subagent which document toolchain to write. Emitted

apps/sim/lib/copilot/chat/post.ts

@@ -169,7 +169,7 @@ type UnifiedChatBranch =
         fileAttachments?: UnifiedChatRequest['fileAttachments']
         userPermission?: string
         userTimezone?: string
-        userMetadata?: { name?: string; timezone?: string }
+        userMetadata?: { name?: string; email?: string; timezone?: string }
         workflowId: string
         workflowName?: string
         workspaceId?: string
@@ -204,7 +204,7 @@ type UnifiedChatBranch =
         fileAttachments?: UnifiedChatRequest['fileAttachments']
         userPermission?: string
         userTimezone?: string
-        userMetadata?: { name?: string; timezone?: string }
+        userMetadata?: { name?: string; email?: string; timezone?: string }
         workspaceContext?: string
       }) => Promise<Record<string, unknown>>
       buildExecutionContext: (params: {
@@ -722,6 +722,7 @@ export async function handleUnifiedChatPost(req: NextRequest) {
     const body = ChatMessageSchema.parse(await req.json())
     const userMetadata = {
       ...(authenticatedUserName ? { name: authenticatedUserName } : {}),
+      ...(authenticatedUserEmail ? { email: authenticatedUserEmail } : {}),
       ...(body.userTimezone ? { timezone: body.userTimezone } : {}),
     }
     const normalizedContexts = normalizeContexts(body.contexts) ?? []

apps/sim/lib/copilot/generated/metrics-v1.ts

@@ -0,0 +1,58 @@
+// AUTO-GENERATED FILE. DO NOT EDIT.
+//
+// Source: copilot/copilot/contracts/metrics-v1.schema.json
+// Regenerate with: bun run metrics-contract:generate
+//
+// Canonical mothership OTel metric names. Call sites should reference
+// `Metric.<Identifier>` (e.g. `Metric.CopilotToolDuration`) rather than raw
+// string literals, so the Go-side contract is the single source of truth and
+// typos become compile errors.
+//
+// NAMES ONLY. Label keys and histogram bucket boundaries are NOT in this
+// contract — Go owns the label-cardinality allowlist and the shared bucket
+// constant, and the Sim emitter MUST mirror those by hand so the Go∪Sim metric
+// union is queryable as one series set.
+
+export const Metric = {
+  CopilotCacheAttempted: 'copilot.cache.attempted',
+  CopilotCacheHit: 'copilot.cache.hit',
+  CopilotCacheWrite: 'copilot.cache.write',
+  CopilotFileReadDuration: 'copilot.file.read.duration',
+  CopilotFileReadSize: 'copilot.file.read.size',
+  CopilotMessagesSerializeDuration: 'copilot.messages.serialize.duration',
+  CopilotRequestCount: 'copilot.request.count',
+  CopilotRequestDuration: 'copilot.request.duration',
+  CopilotToolCalls: 'copilot.tool.calls',
+  CopilotToolDuration: 'copilot.tool.duration',
+  CopilotVfsMaterializeDuration: 'copilot.vfs.materialize.duration',
+  GenAiClientCacheTokenUsage: 'gen_ai.client.cache.token.usage',
+  GenAiClientTokenUsage: 'gen_ai.client.token.usage',
+  LlmClientErrors: 'llm.client.errors',
+  LlmClientOutputCutoff: 'llm.client.output_cutoff',
+  LlmClientStreamDuration: 'llm.client.stream.duration',
+  LlmClientTimeToFirstToken: 'llm.client.time_to_first_token',
+} as const
+
+export type MetricKey = keyof typeof Metric
+export type MetricValue = (typeof Metric)[MetricKey]
+
+/** Readonly sorted list of every canonical mothership metric name. */
+export const MetricValues: readonly MetricValue[] = [
+  'copilot.cache.attempted',
+  'copilot.cache.hit',
+  'copilot.cache.write',
+  'copilot.file.read.duration',
+  'copilot.file.read.size',
+  'copilot.messages.serialize.duration',
+  'copilot.request.count',
+  'copilot.request.duration',
+  'copilot.tool.calls',
+  'copilot.tool.duration',
+  'copilot.vfs.materialize.duration',
+  'gen_ai.client.cache.token.usage',
+  'gen_ai.client.token.usage',
+  'llm.client.errors',
+  'llm.client.output_cutoff',
+  'llm.client.stream.duration',
+  'llm.client.time_to_first_token',
+] as const