Use Claude Code with your ChatGPT, Kimi, Cursor or Grok subscription via a local Anthropic-compatible proxy
Rust
add per-session compaction timeline telemetry
Raine Virta952da85a6b4e
diff --git a/src/server.ts b/src/server.tsindex 1b68f2b..b0e0e07 100644--- a/src/server.ts+++ b/src/server.ts@@ -14,6 +14,45 @@ export interface ServeOptions {port: number}+interface SessionCountSnapshot {+ reqId: string+ model: string+ messageCount: number+ toolCount: number+ tokens: number+}++interface SessionMessageSnapshot {+ reqId: string+ model: string+ messageCount: number+ toolCount: number+ localInputTokens?: number+ translatedInputTokens?: number+}++interface SessionTimelineState {+ seq: number+ lastCount?: SessionCountSnapshot+ lastMessage?: SessionMessageSnapshot+}++const sessionTimeline = new Map<string, SessionTimelineState>()++function nextSessionTimeline(sessionId?: string): {+ state?: SessionTimelineState+ sessionSeq?: number+} {+ if (!sessionId) return {}+ let state = sessionTimeline.get(sessionId)+ if (!state) {+ state = { seq: 0 }+ sessionTimeline.set(sessionId, state)+ }+ state.seq += 1+ return { state, sessionSeq: state.seq }+}+export function startServer(opts: ServeOptions): { stop: () => void; port: number } {const server = Bun.serve({hostname: "127.0.0.1",@@ -51,15 +90,37 @@ async function route(req: Request, url: URL, reqId: string): Promise<Response> {if (req.method === "POST" && url.pathname === "/v1/messages/count_tokens") {const body = (await req.json()) as AnthropicRequestconst tokens = countTokens(body)+ const sessionId = req.headers.get("x-claude-code-session-id") || undefined+ const { state, sessionSeq } = nextSessionTimeline(sessionId)+ const messageCount = body.messages?.length ?? 0+ const toolCount = body.tools?.length ?? 0log.debug("count_tokens", { reqId, tokens })+ if (state) {+ state.lastCount = {+ reqId,+ model: body.model,+ messageCount,+ toolCount,+ tokens,+ }+ }if (VERBOSE) {log.info("compaction telemetry", {reqId,+ phase: "count_tokens",path: url.pathname,+ sessionId,+ sessionSeq,model: body.model,tokens,- messageCount: body.messages?.length ?? 0,- toolCount: body.tools?.length ?? 0,+ messageCount,+ toolCount,+ previousMessageReqId: state?.lastMessage?.reqId,+ previousMessageModel: state?.lastMessage?.model,+ previousMessageCount: state?.lastMessage?.messageCount,+ previousMessageToolCount: state?.lastMessage?.toolCount,+ previousMessageLocalInputTokens: state?.lastMessage?.localInputTokens,+ previousMessageTranslatedInputTokens: state?.lastMessage?.translatedInputTokens,})}return new Response(JSON.stringify({ input_tokens: tokens }), {@@ -83,14 +144,17 @@ async function handleMessages(req: Request, reqId: string): Promise<Response> {}const sessionId = req.headers.get("x-claude-code-session-id") || undefined+ const { state, sessionSeq } = nextSessionTimeline(sessionId)const messageId = `msg_${crypto.randomUUID().replace(/-/g, "")}`const wantStream = body.stream !== false+ const messageCount = body.messages?.length ?? 0+ const toolCount = body.tools?.length ?? 0log.debug("anthropic request", {reqId,model: body.model,- messageCount: body.messages?.length,- toolCount: body.tools?.length ?? 0,+ messageCount,+ toolCount,stream: wantStream,sessionId,hasJsonSchemaFormat: body.output_config?.format?.type === "json_schema",@@ -115,6 +179,16 @@ async function handleMessages(req: Request, reqId: string): Promise<Response> {const translated = translateRequest({ ...body, model: resolvedModel }, { sessionId })const localInputTokens = VERBOSE ? countTokens(body) : undefinedconst translatedInputTokens = VERBOSE ? countTranslatedTokens(translated) : undefined+ if (state) {+ state.lastMessage = {+ reqId,+ model: body.model,+ messageCount,+ toolCount,+ localInputTokens,+ translatedInputTokens,+ }+ }log.debug("translated request", {reqId,requestedModel: body.model,@@ -129,14 +203,22 @@ async function handleMessages(req: Request, reqId: string): Promise<Response> {log.info("compaction telemetry", {reqId,phase: "translated_request",+ sessionId,+ sessionSeq,requestedModel: body.model,resolvedModel,+ messageCount,+ toolCount,localInputTokens,translatedInputTokens,inputItems: translated.input.length,- toolCount: translated.tools?.length ?? 0,+ translatedToolCount: translated.tools?.length ?? 0,hasInstructions: !!translated.instructions,- sessionId,+ previousCountReqId: state?.lastCount?.reqId,+ previousCountModel: state?.lastCount?.model,+ previousCountTokens: state?.lastCount?.tokens,+ previousCountMessageCount: state?.lastCount?.messageCount,+ previousCountToolCount: state?.lastCount?.toolCount,})}@@ -176,15 +258,18 @@ async function handleMessages(req: Request, reqId: string): Promise<Response> {reqId,phase: "upstream_finish",mode: "stream",+ sessionId,+ sessionSeq,requestedModel: body.model,resolvedModel,+ messageCount,+ toolCount,localInputTokens,translatedInputTokens,upstreamInputTokens: finish.usage?.input_tokens ?? 0,upstreamOutputTokens: finish.usage?.output_tokens ?? 0,upstreamCachedInputTokens: finish.usage?.input_tokens_details?.cached_tokens ?? 0,stopReason: finish.stopReason,- sessionId,})}: undefined,@@ -206,15 +291,18 @@ async function handleMessages(req: Request, reqId: string): Promise<Response> {reqId,phase: "upstream_finish",mode: "non_stream",+ sessionId,+ sessionSeq,requestedModel: body.model,resolvedModel,+ messageCount,+ toolCount,localInputTokens,translatedInputTokens,upstreamInputTokens: result.usage.input_tokens,upstreamOutputTokens: result.usage.output_tokens,upstreamCachedInputTokens: result.usage.cache_read_input_tokens,stopReason: result.stop_reason,- sessionId,})}return new Response(JSON.stringify(result), {