@curatedmcp/tokenshield-core 0.2.0

package/dist/server.js

@@ -0,0 +1,130 @@
+import { createServer } from "node:http";
+import { handleAnthropicRequest } from "./proxy/anthropic-passthrough.js";
+import { Ledger } from "./ledger.js";
+export function defaultConfig(overrides = {}) {
+    const home = process.env["HOME"] ?? process.env["USERPROFILE"] ?? ".";
+    return {
+        upstreamBaseUrl: overrides.upstreamBaseUrl ?? "https://api.anthropic.com",
+        port: overrides.port ?? 7777,
+        bind: overrides.bind ?? "127.0.0.1",
+        dashboardPort: overrides.dashboardPort ?? 7778,
+        ledgerPath: overrides.ledgerPath ?? `${home}/.tokenshield/ledger.db`,
+        enabledProcessors: overrides.enabledProcessors ?? ["token-accounting"],
+        retentionDays: overrides.retentionDays ?? 7,
+    };
+}
+async function listenOn(server, port, bind) {
+    return new Promise((resolve, reject) => {
+        server.once("error", reject);
+        server.listen(port, bind, () => {
+            server.removeListener("error", reject);
+            resolve();
+        });
+    });
+}
+async function closeServer(server) {
+    return new Promise((resolve) => {
+        server.close(() => resolve());
+    });
+}
+export async function start(opts) {
+    const ledger = new Ledger(opts.config.ledgerPath);
+    const sink = (r) => {
+        try {
+            ledger.record(r);
+        }
+        catch {
+            // ledger errors must never break the request path
+        }
+        opts.onRecord?.(r);
+    };
+    const proxy = createServer((req, res) => {
+        if (req.url === "/__tokenshield/health") {
+            res.statusCode = 200;
+            res.setHeader("content-type", "application/json");
+            res.end(JSON.stringify({ ok: true, version: "0.1.0" }));
+            return;
+        }
+        handleAnthropicRequest(req, res, opts.config, sink).catch((err) => {
+            if (!res.headersSent) {
+                res.statusCode = 500;
+                res.setHeader("content-type", "application/json");
+                res.end(JSON.stringify({
+                    type: "error",
+                    error: {
+                        type: "tokenshield_internal_error",
+                        message: err?.message ?? "unknown",
+                    },
+                }));
+            }
+            else {
+                try {
+                    res.end();
+                }
+                catch {
+                    // ignore
+                }
+            }
+        });
+    });
+    proxy.keepAliveTimeout = 65_000;
+    proxy.headersTimeout = 70_000;
+    proxy.requestTimeout = 0; // streaming responses can be long
+    const dashboard = createServer((req, res) => {
+        const url = req.url ?? "/";
+        if (url === "/api/summary") {
+            const since = Date.now() - 24 * 60 * 60 * 1000;
+            const summary = ledger.summary(since);
+            res.statusCode = 200;
+            res.setHeader("content-type", "application/json");
+            res.setHeader("cache-control", "no-store");
+            res.end(JSON.stringify(summary));
+            return;
+        }
+        if (url === "/api/recent") {
+            res.statusCode = 200;
+            res.setHeader("content-type", "application/json");
+            res.setHeader("cache-control", "no-store");
+            res.end(JSON.stringify(ledger.recent(50)));
+            return;
+        }
+        if (url === "/health") {
+            res.statusCode = 200;
+            res.setHeader("content-type", "application/json");
+            res.end(JSON.stringify({ ok: true }));
+            return;
+        }
+        const html = opts.renderDashboard?.(ledger) ?? defaultDashboardHtml();
+        res.statusCode = 200;
+        res.setHeader("content-type", "text/html; charset=utf-8");
+        res.setHeader("cache-control", "no-store");
+        res.end(html);
+    });
+    await listenOn(proxy, opts.config.port, opts.config.bind);
+    await listenOn(dashboard, opts.config.dashboardPort, opts.config.bind);
+    const retentionInterval = setInterval(() => {
+        const cutoff = Date.now() - opts.config.retentionDays * 24 * 60 * 60 * 1000;
+        try {
+            ledger.prune(cutoff);
+        }
+        catch {
+            // ignore
+        }
+    }, 60 * 60 * 1000);
+    retentionInterval.unref();
+    return {
+        proxy,
+        dashboard,
+        ledger,
+        close: async () => {
+            clearInterval(retentionInterval);
+            await Promise.all([closeServer(proxy), closeServer(dashboard)]);
+            ledger.close();
+        },
+    };
+}
+function defaultDashboardHtml() {
+    return `<!doctype html><meta charset="utf-8"><title>TokenShield</title>
+<body><h1>TokenShield</h1><p>Dashboard renderer not provided.</p></body>`;
+}
+//# sourceMappingURL=server.js.map

package/dist/server.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"server.js","sourceRoot":"","sources":["../src/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAA2C,MAAM,WAAW,CAAC;AAElF,OAAO,EAAE,sBAAsB,EAAE,MAAM,kCAAkC,CAAC;AAC1E,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AAiBrC,MAAM,UAAU,aAAa,CAAC,YAAkC,EAAE;IAChE,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,IAAI,GAAG,CAAC;IACtE,OAAO;QACL,eAAe,EAAE,SAAS,CAAC,eAAe,IAAI,2BAA2B;QACzE,IAAI,EAAE,SAAS,CAAC,IAAI,IAAI,IAAI;QAC5B,IAAI,EAAE,SAAS,CAAC,IAAI,IAAI,WAAW;QACnC,aAAa,EAAE,SAAS,CAAC,aAAa,IAAI,IAAI;QAC9C,UAAU,EAAE,SAAS,CAAC,UAAU,IAAI,GAAG,IAAI,yBAAyB;QACpE,iBAAiB,EAAE,SAAS,CAAC,iBAAiB,IAAI,CAAC,kBAAkB,CAAC;QACtE,aAAa,EAAE,SAAS,CAAC,aAAa,IAAI,CAAC;KAC5C,CAAC;AACJ,CAAC;AAED,KAAK,UAAU,QAAQ,CAAC,MAAc,EAAE,IAAY,EAAE,IAAY;IAChE,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;QAC7B,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,EAAE,GAAG,EAAE;YAC7B,MAAM,CAAC,cAAc,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YACvC,OAAO,EAAE,CAAC;QACZ,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,MAAc;IACvC,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE;QAC7B,MAAM,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,IAAkB;IAC5C,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC;IAElD,MAAM,IAAI,GAAG,CAAC,CAAgB,EAAQ,EAAE;QACtC,IAAI,CAAC;YACH,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACnB,CAAC;QAAC,MAAM,CAAC;YACP,kDAAkD;QACpD,CAAC;QACD,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC;IACrB,CAAC,CAAC;IAEF,MAAM,KAAK,GAAG,YAAY,CAAC,CAAC,GAAoB,EAAE,GAAmB,EAAE,EAAE;QACvE,IAAI,GAAG,CAAC,GAAG,KAAK,uBAAuB,EAAE,CAAC;YACxC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;YACrB,GAAG,CAAC,SAAS,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;YAClD,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC;YACxD,OAAO;QACT,CAAC;QACD,sBAAsB,CAAC,GAAG,EAAE,GAAG,EAAE,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,GAAY,EAAE,EAAE;YACzE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,CAAC;gBACrB,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;gBACrB,GAAG,CAAC,SAAS,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;gBAClD,GAAG,CAAC,GAAG,CACL,IAAI,CAAC,SAAS,CAAC;oBACb,IAAI,EAAE,OAAO;oBACb,KAAK,EAAE;wBACL,IAAI,EAAE,4BAA4B;wBAClC,OAAO,EAAG,GAAa,EAAE,OAAO,IAAI,SAAS;qBAC9C;iBACF,CAAC,CACH,CAAC;YACJ,CAAC;iBAAM,CAAC;gBACN,IAAI,CAAC;oBACH,GAAG,CAAC,GAAG,EAAE,CAAC;gBACZ,CAAC;gBAAC,MAAM,CAAC;oBACP,SAAS;gBACX,CAAC;YACH,CAAC;QACH,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IACH,KAAK,CAAC,gBAAgB,GAAG,MAAM,CAAC;IAChC,KAAK,CAAC,cAAc,GAAG,MAAM,CAAC;IAC9B,KAAK,CAAC,cAAc,GAAG,CAAC,CAAC,CAAC,kCAAkC;IAE5D,MAAM,SAAS,GAAG,YAAY,CAAC,CAAC,GAAoB,EAAE,GAAmB,EAAE,EAAE;QAC3E,MAAM,GAAG,GAAG,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC;QAC3B,IAAI,GAAG,KAAK,cAAc,EAAE,CAAC;YAC3B,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;YAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YACtC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;YACrB,GAAG,CAAC,SAAS,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;YAClD,GAAG,CAAC,SAAS,CAAC,eAAe,EAAE,UAAU,CAAC,CAAC;YAC3C,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;YACjC,OAAO;QACT,CAAC;QACD,IAAI,GAAG,KAAK,aAAa,EAAE,CAAC;YAC1B,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;YACrB,GAAG,CAAC,SAAS,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;YAClD,GAAG,CAAC,SAAS,CAAC,eAAe,EAAE,UAAU,CAAC,CAAC;YAC3C,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;YAC3C,OAAO;QACT,CAAC;QACD,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;YACtB,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;YACrB,GAAG,CAAC,SAAS,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;YAClD,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;YACtC,OAAO;QACT,CAAC;QACD,MAAM,IAAI,GAAG,IAAI,CAAC,eAAe,EAAE,CAAC,MAAM,CAAC,IAAI,oBAAoB,EAAE,CAAC;QACtE,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QACrB,GAAG,CAAC,SAAS,CAAC,cAAc,EAAE,0BAA0B,CAAC,CAAC;QAC1D,GAAG,CAAC,SAAS,CAAC,eAAe,EAAE,UAAU,CAAC,CAAC;QAC3C,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAChB,CAAC,CAAC,CAAC;IAEH,MAAM,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAC1D,MAAM,QAAQ,CAAC,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAEvE,MAAM,iBAAiB,GAAG,WAAW,CAAC,GAAG,EAAE;QACzC,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,MAAM,CAAC,aAAa,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;QAC5E,IAAI,CAAC;YACH,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACvB,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;IACH,CAAC,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC;IACnB,iBAAiB,CAAC,KAAK,EAAE,CAAC;IAE1B,OAAO;QACL,KAAK;QACL,SAAS;QACT,MAAM;QACN,KAAK,EAAE,KAAK,IAAI,EAAE;YAChB,aAAa,CAAC,iBAAiB,CAAC,CAAC;YACjC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,WAAW,CAAC,KAAK,CAAC,EAAE,WAAW,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;YAChE,MAAM,CAAC,KAAK,EAAE,CAAC;QACjB,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,oBAAoB;IAC3B,OAAO;yEACgE,CAAC;AAC1E,CAAC"}

package/dist/types.d.ts

@@ -0,0 +1,36 @@
+export type ModelId = "claude-opus-4-7" | "claude-opus-4-6" | "claude-sonnet-4-6" | "claude-sonnet-4-5" | "claude-haiku-4-5" | "claude-haiku-4-5-20251001" | string;
+export interface UsageCounts {
+    inputTokens: number;
+    outputTokens: number;
+    cacheCreationInputTokens: number;
+    cacheReadInputTokens: number;
+}
+export interface RequestRecord {
+    id: string;
+    timestamp: number;
+    model: ModelId;
+    endpoint: string;
+    streamed: boolean;
+    durationMs: number;
+    upstreamStatus: number;
+    upstreamError: string | null;
+    usageRaw: UsageCounts;
+    usageSent: UsageCounts;
+    dollarsRaw: number;
+    dollarsSent: number;
+    dollarsSaved: number;
+    processorsApplied: string[];
+}
+export interface ProxyConfig {
+    upstreamBaseUrl: string;
+    port: number;
+    bind: string;
+    dashboardPort: number;
+    ledgerPath: string;
+    enabledProcessors: string[];
+    retentionDays: number;
+}
+export type SSEEvent = {
+    event: string;
+    data: string;
+};

package/dist/types.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=types.js.map

package/dist/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":""}

package/package.json

@@ -0,0 +1,38 @@
+{
+  "name": "@curatedmcp/tokenshield-core",
+  "version": "0.2.0",
+  "description": "TokenShield proxy engine — Anthropic API-layer middleware with token accounting",
+  "type": "module",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": "./dist/index.js",
+    "./server": "./dist/server.js",
+    "./pipeline": "./dist/proxy/pipeline.js",
+    "./ledger": "./dist/ledger.js",
+    "./pricing": "./dist/pricing.js"
+  },
+  "files": [
+    "dist",
+    "src"
+  ],
+  "scripts": {
+    "build": "tsc",
+    "dev": "tsc --watch",
+    "test": "tsc -p tsconfig.test.json && node --enable-source-maps --test $(find dist-test/test -name '*.test.js')",
+    "prepublishOnly": "npm run build"
+  },
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/oneprofile-dev/tokenshield"
+  },
+  "license": "MIT",
+  "dependencies": {},
+  "devDependencies": {
+    "@types/node": "^22.0.0",
+    "typescript": "^5.5.0"
+  },
+  "engines": {
+    "node": ">=22.0.0"
+  }
+}

package/src/index.ts

@@ -0,0 +1,31 @@
+export { start, defaultConfig } from "./server.js";
+export type { ProxyServerHandle, StartOptions } from "./server.js";
+export { Ledger } from "./ledger.js";
+export type { SavingsSummary } from "./ledger.js";
+export { dollarsFor, addUsage, emptyUsage, isKnownModel } from "./pricing.js";
+export { SSEParser } from "./proxy/sse.js";
+export { StreamUsageAccumulator, usageFromJson } from "./proxy/usage.js";
+export {
+  handleAnthropicRequest,
+  setProcessorEnabled,
+  getProcessorEnabledIds,
+  getResponseCacheStats,
+} from "./proxy/anthropic-passthrough.js";
+export { providerForPath, anthropic } from "./providers/registry.js";
+export type { Provider, Conversation, ConvMessage, ConvBlock, ProviderId } from "./providers/types.js";
+export { Pipeline } from "./processors/pipeline.js";
+export { conversationDedup } from "./processors/conversation-dedup.js";
+export { ResponseCache } from "./processors/response-cache.js";
+export type {
+  Processor,
+  ProcessorContext,
+  ProcessorEffect,
+  ProcessorResult,
+} from "./processors/types.js";
+export type {
+  RequestRecord,
+  ProxyConfig,
+  UsageCounts,
+  ModelId,
+  SSEEvent,
+} from "./types.js";

package/src/ledger.ts

@@ -0,0 +1,232 @@
+import { DatabaseSync } from "node:sqlite";
+import { mkdirSync } from "node:fs";
+import { dirname } from "node:path";
+import type { RequestRecord } from "./types.js";
+
+export interface SavingsSummary {
+  windowStart: number;
+  windowEnd: number;
+  requestCount: number;
+  totalInputTokensRaw: number;
+  totalInputTokensSent: number;
+  totalOutputTokensRaw: number;
+  totalOutputTokensSent: number;
+  dollarsRaw: number;
+  dollarsSent: number;
+  dollarsSaved: number;
+  byModel: Array<{
+    model: string;
+    requests: number;
+    inputTokens: number;
+    outputTokens: number;
+    dollars: number;
+  }>;
+}
+
+interface SummaryRow {
+  model: string;
+  requests: number;
+  input_tokens_raw: number | null;
+  input_tokens_sent: number | null;
+  output_tokens_raw: number | null;
+  output_tokens_sent: number | null;
+  dollars_raw: number | null;
+  dollars_sent: number | null;
+  dollars_saved: number | null;
+}
+
+interface RequestRow {
+  id: string;
+  timestamp: number;
+  model: string;
+  endpoint: string;
+  streamed: number;
+  duration_ms: number;
+  upstream_status: number;
+  upstream_error: string | null;
+  input_tokens_raw: number;
+  input_tokens_sent: number;
+  output_tokens_raw: number;
+  output_tokens_sent: number;
+  cache_create_raw: number;
+  cache_read_raw: number;
+  dollars_raw: number;
+  dollars_sent: number;
+  dollars_saved: number;
+  processors: string;
+}
+
+export class Ledger {
+  private db: DatabaseSync;
+  private insertStmt: ReturnType<DatabaseSync["prepare"]>;
+  private summaryStmt: ReturnType<DatabaseSync["prepare"]>;
+  private recentStmt: ReturnType<DatabaseSync["prepare"]>;
+  private pruneStmt: ReturnType<DatabaseSync["prepare"]>;
+
+  constructor(path: string) {
+    mkdirSync(dirname(path), { recursive: true });
+    this.db = new DatabaseSync(path);
+    this.db.exec("PRAGMA journal_mode = WAL");
+    this.db.exec("PRAGMA synchronous = NORMAL");
+    this.db.exec(`
+      CREATE TABLE IF NOT EXISTS requests (
+        id TEXT PRIMARY KEY,
+        timestamp INTEGER NOT NULL,
+        model TEXT NOT NULL,
+        endpoint TEXT NOT NULL,
+        streamed INTEGER NOT NULL,
+        duration_ms INTEGER NOT NULL,
+        upstream_status INTEGER NOT NULL,
+        upstream_error TEXT,
+        input_tokens_raw INTEGER NOT NULL,
+        input_tokens_sent INTEGER NOT NULL,
+        output_tokens_raw INTEGER NOT NULL,
+        output_tokens_sent INTEGER NOT NULL,
+        cache_create_raw INTEGER NOT NULL,
+        cache_read_raw INTEGER NOT NULL,
+        dollars_raw REAL NOT NULL,
+        dollars_sent REAL NOT NULL,
+        dollars_saved REAL NOT NULL,
+        processors TEXT NOT NULL
+      );
+    `);
+    this.db.exec("CREATE INDEX IF NOT EXISTS idx_requests_ts ON requests(timestamp)");
+    this.db.exec("CREATE INDEX IF NOT EXISTS idx_requests_model ON requests(model)");
+
+    this.insertStmt = this.db.prepare(`
+      INSERT INTO requests (
+        id, timestamp, model, endpoint, streamed, duration_ms,
+        upstream_status, upstream_error,
+        input_tokens_raw, input_tokens_sent,
+        output_tokens_raw, output_tokens_sent,
+        cache_create_raw, cache_read_raw,
+        dollars_raw, dollars_sent, dollars_saved, processors
+      ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+    `);
+    this.summaryStmt = this.db.prepare(`
+      SELECT model,
+             COUNT(*) as requests,
+             SUM(input_tokens_raw)  as input_tokens_raw,
+             SUM(input_tokens_sent) as input_tokens_sent,
+             SUM(output_tokens_raw)  as output_tokens_raw,
+             SUM(output_tokens_sent) as output_tokens_sent,
+             SUM(dollars_raw)   as dollars_raw,
+             SUM(dollars_sent)  as dollars_sent,
+             SUM(dollars_saved) as dollars_saved
+        FROM requests
+       WHERE timestamp >= ?
+       GROUP BY model
+       ORDER BY dollars_raw DESC
+    `);
+    this.recentStmt = this.db.prepare(
+      `SELECT * FROM requests ORDER BY timestamp DESC LIMIT ?`,
+    );
+    this.pruneStmt = this.db.prepare(`DELETE FROM requests WHERE timestamp < ?`);
+  }
+
+  record(r: RequestRecord): void {
+    this.insertStmt.run(
+      r.id,
+      r.timestamp,
+      r.model,
+      r.endpoint,
+      r.streamed ? 1 : 0,
+      r.durationMs,
+      r.upstreamStatus,
+      r.upstreamError,
+      r.usageRaw.inputTokens,
+      r.usageSent.inputTokens,
+      r.usageRaw.outputTokens,
+      r.usageSent.outputTokens,
+      r.usageRaw.cacheCreationInputTokens,
+      r.usageRaw.cacheReadInputTokens,
+      r.dollarsRaw,
+      r.dollarsSent,
+      r.dollarsSaved,
+      JSON.stringify(r.processorsApplied),
+    );
+  }
+
+  summary(sinceMs: number): SavingsSummary {
+    const now = Date.now();
+    const rows = this.summaryStmt.all(sinceMs) as unknown as SummaryRow[];
+
+    let totalIRaw = 0,
+      totalISent = 0,
+      totalORaw = 0,
+      totalOSent = 0,
+      dRaw = 0,
+      dSent = 0,
+      dSaved = 0,
+      reqCount = 0;
+    for (const r of rows) {
+      reqCount += r.requests;
+      totalIRaw += r.input_tokens_raw ?? 0;
+      totalISent += r.input_tokens_sent ?? 0;
+      totalORaw += r.output_tokens_raw ?? 0;
+      totalOSent += r.output_tokens_sent ?? 0;
+      dRaw += r.dollars_raw ?? 0;
+      dSent += r.dollars_sent ?? 0;
+      dSaved += r.dollars_saved ?? 0;
+    }
+
+    return {
+      windowStart: sinceMs,
+      windowEnd: now,
+      requestCount: reqCount,
+      totalInputTokensRaw: totalIRaw,
+      totalInputTokensSent: totalISent,
+      totalOutputTokensRaw: totalORaw,
+      totalOutputTokensSent: totalOSent,
+      dollarsRaw: dRaw,
+      dollarsSent: dSent,
+      dollarsSaved: dSaved,
+      byModel: rows.map((r) => ({
+        model: r.model,
+        requests: r.requests,
+        inputTokens: r.input_tokens_raw ?? 0,
+        outputTokens: r.output_tokens_raw ?? 0,
+        dollars: r.dollars_raw ?? 0,
+      })),
+    };
+  }
+
+  recent(limit = 50): RequestRecord[] {
+    const rows = this.recentStmt.all(limit) as unknown as RequestRow[];
+    return rows.map((r) => ({
+      id: r.id,
+      timestamp: r.timestamp,
+      model: r.model,
+      endpoint: r.endpoint,
+      streamed: r.streamed === 1,
+      durationMs: r.duration_ms,
+      upstreamStatus: r.upstream_status,
+      upstreamError: r.upstream_error,
+      usageRaw: {
+        inputTokens: r.input_tokens_raw,
+        outputTokens: r.output_tokens_raw,
+        cacheCreationInputTokens: r.cache_create_raw,
+        cacheReadInputTokens: r.cache_read_raw,
+      },
+      usageSent: {
+        inputTokens: r.input_tokens_sent,
+        outputTokens: r.output_tokens_sent,
+        cacheCreationInputTokens: r.cache_create_raw,
+        cacheReadInputTokens: r.cache_read_raw,
+      },
+      dollarsRaw: r.dollars_raw,
+      dollarsSent: r.dollars_sent,
+      dollarsSaved: r.dollars_saved,
+      processorsApplied: JSON.parse(r.processors ?? "[]") as string[],
+    }));
+  }
+
+  prune(olderThanMs: number): number {
+    const result = this.pruneStmt.run(olderThanMs);
+    return Number(result.changes);
+  }
+
+  close(): void {
+    this.db.close();
+  }
+}

package/src/pricing.ts

@@ -0,0 +1,93 @@
+import type { ModelId, UsageCounts } from "./types.js";
+
+interface ModelPricing {
+  inputPerMTok: number;
+  outputPerMTok: number;
+  cacheWritePerMTok: number;
+  cacheReadPerMTok: number;
+}
+
+const PRICING: Record<string, ModelPricing> = {
+  "claude-opus-4-7": {
+    inputPerMTok: 15,
+    outputPerMTok: 75,
+    cacheWritePerMTok: 18.75,
+    cacheReadPerMTok: 1.5,
+  },
+  "claude-opus-4-6": {
+    inputPerMTok: 15,
+    outputPerMTok: 75,
+    cacheWritePerMTok: 18.75,
+    cacheReadPerMTok: 1.5,
+  },
+  "claude-sonnet-4-6": {
+    inputPerMTok: 3,
+    outputPerMTok: 15,
+    cacheWritePerMTok: 3.75,
+    cacheReadPerMTok: 0.3,
+  },
+  "claude-sonnet-4-5": {
+    inputPerMTok: 3,
+    outputPerMTok: 15,
+    cacheWritePerMTok: 3.75,
+    cacheReadPerMTok: 0.3,
+  },
+  "claude-haiku-4-5": {
+    inputPerMTok: 1,
+    outputPerMTok: 5,
+    cacheWritePerMTok: 1.25,
+    cacheReadPerMTok: 0.1,
+  },
+};
+
+const DEFAULT_PRICING: ModelPricing = PRICING["claude-sonnet-4-6"]!;
+
+function lookup(model: ModelId): ModelPricing {
+  const exact = PRICING[model];
+  if (exact) return exact;
+  for (const [key, val] of Object.entries(PRICING)) {
+    if (model.startsWith(key)) return val;
+  }
+  const lower = model.toLowerCase();
+  if (lower.includes("opus")) return PRICING["claude-opus-4-7"]!;
+  if (lower.includes("haiku")) return PRICING["claude-haiku-4-5"]!;
+  if (lower.includes("sonnet")) return PRICING["claude-sonnet-4-6"]!;
+  return DEFAULT_PRICING;
+}
+
+export function dollarsFor(model: ModelId, u: UsageCounts): number {
+  const p = lookup(model);
+  return (
+    (u.inputTokens * p.inputPerMTok) / 1_000_000 +
+    (u.outputTokens * p.outputPerMTok) / 1_000_000 +
+    (u.cacheCreationInputTokens * p.cacheWritePerMTok) / 1_000_000 +
+    (u.cacheReadInputTokens * p.cacheReadPerMTok) / 1_000_000
+  );
+}
+
+export function emptyUsage(): UsageCounts {
+  return {
+    inputTokens: 0,
+    outputTokens: 0,
+    cacheCreationInputTokens: 0,
+    cacheReadInputTokens: 0,
+  };
+}
+
+export function addUsage(a: UsageCounts, b: UsageCounts): UsageCounts {
+  return {
+    inputTokens: a.inputTokens + b.inputTokens,
+    outputTokens: a.outputTokens + b.outputTokens,
+    cacheCreationInputTokens:
+      a.cacheCreationInputTokens + b.cacheCreationInputTokens,
+    cacheReadInputTokens: a.cacheReadInputTokens + b.cacheReadInputTokens,
+  };
+}
+
+export function isKnownModel(model: ModelId): boolean {
+  if (PRICING[model]) return true;
+  for (const key of Object.keys(PRICING)) {
+    if (model.startsWith(key)) return true;
+  }
+  return false;
+}

package/src/processors/conversation-dedup.ts

@@ -0,0 +1,77 @@
+import type { Conversation, ConvBlock } from "../providers/types.js";
+import type { Processor, ProcessorContext, ProcessorResult } from "./types.js";
+
+/**
+ * Conversation deduplication.
+ *
+ * Walks the conversation in message order. The first occurrence of any
+ * tool_result content (keyed by its content hash) is kept verbatim. Subsequent
+ * occurrences are replaced with a compact pointer that references the prior one.
+ *
+ * Determinism: same input always produces the same output (hash-based ordering),
+ * so Anthropic prompt caching remains valid.
+ *
+ * Fail-open: if any block can't be handled cleanly, it's left untouched.
+ */
+class ConversationDedup implements Processor {
+  readonly id = "conversation-dedup";
+  readonly enabledByDefault = true;
+
+  /** Only elide payloads worth eliding — avoids stub-overhead on tiny results. */
+  static readonly MIN_ELIDE_BYTES = 256;
+
+  onRequest(conv: Conversation, _ctx: ProcessorContext): ProcessorResult {
+    const seen = new Map<string, { messageIndex: number; toolUseId: string }>();
+    let elidedBytesTotal = 0;
+    let elidedCount = 0;
+    let mutated = false;
+
+    const newMessages = conv.messages.map((msg, msgIdx) => {
+      const newBlocks = msg.blocks.map<ConvBlock>((block) => {
+        if (block.kind !== "tool_result") return block;
+        if (block.pointer) return block; // already deduped (idempotency)
+        const prior = seen.get(block.contentHash);
+        if (prior === undefined) {
+          seen.set(block.contentHash, {
+            messageIndex: msgIdx,
+            toolUseId: block.tool_use_id,
+          });
+          return block;
+        }
+        if (block.contentBytes < ConversationDedup.MIN_ELIDE_BYTES) {
+          // Not worth the stub overhead; leave verbatim.
+          return block;
+        }
+        mutated = true;
+        elidedBytesTotal += block.contentBytes;
+        elidedCount++;
+        return {
+          ...block,
+          pointer: {
+            priorMessageIndex: prior.messageIndex,
+            priorToolUseId: prior.toolUseId,
+            elidedBytes: block.contentBytes,
+          },
+        };
+      });
+      return mutated ? { ...msg, blocks: newBlocks } : msg;
+    });
+
+    if (!mutated) {
+      return { conversation: conv, effects: [] };
+    }
+
+    return {
+      conversation: { ...conv, messages: newMessages },
+      effects: [
+        {
+          name: this.id,
+          bytesSaved: elidedBytesTotal,
+          detail: { elidedCount, distinctContent: seen.size },
+        },
+      ],
+    };
+  }
+}
+
+export const conversationDedup = new ConversationDedup();