codeep 2.13.2 → 2.15.0

package/dist/utils/tokenTracker.d.ts

@@ -46,6 +46,18 @@ export declare function getPricingTable(): {
     inputPer1M: number;
     outputPer1M: number;
 }[];
+/** An isolated token-record buffer for one scope (e.g. one ACP session). */
+export type TokenScope = TokenRecord[];
+/** Create a fresh, empty scope buffer (one per ACP session). */
+export declare function createTokenScope(): TokenScope;
+/**
+ * Run `fn` with `scope` as the active token-record buffer. Every
+ * recordTokenUsage() call made within `fn`'s async flow (including across
+ * awaits) accumulates into `scope`, and reads (getCostBreakdown/…) made in the
+ * same flow see only `scope`. Used by the ACP server to isolate per-session
+ * usage without threading a session id through the deep API layer.
+ */
+export declare function runWithTokenScope<T>(scope: TokenScope, fn: () => T): T;
 /**
  * Record token usage from an API response. The optional `actualCostUsd`
  * argument lets aggregator providers (OpenRouter) pass through the
@@ -76,9 +88,14 @@ export interface ProviderCostBreakdown {
     estimatedCost: number;
 }
 /**
- * Get cost breakdown grouped by provider/model
+ * Get cost breakdown grouped by provider/model.
+ *
+ * `startIndex` lets callers price only the records appended since a marker (see
+ * getRecordCount) — used to report a single run/prompt's delta to cloud
+ * telemetry WITHOUT wiping the session-cumulative store the status bar and
+ * `/cost` read. Defaults to 0 (the whole current scope).
  */
-export declare function getCostBreakdown(): ProviderCostBreakdown[];
+export declare function getCostBreakdown(startIndex?: number): ProviderCostBreakdown[];
 /**
  * Aggregate Anthropic prompt-caching stats for the current session.
  * Returns the breakdown plus an estimate of what the input billing would
@@ -105,7 +122,17 @@ export declare function getLastUsage(): TokenRecord | null;
  */
 export declare function formatTokenCount(tokens: number): string;
 /**
- * Reset session tracking
+ * Number of records in the current scope. Capture before a run/prompt and pass
+ * it to getCostBreakdown(startIndex) to price just that run's delta (for cloud
+ * telemetry) without wiping the cumulative store the status bar and `/cost`
+ * read.
+ */
+export declare function getRecordCount(): number;
+/**
+ * Reset the current scope's tracking. Production run/prompt paths no longer
+ * call this (they use getRecordCount + getCostBreakdown(startIndex) so the
+ * session-cumulative totals survive); retained for the test suite, which uses
+ * it to isolate the process-wide default buffer between cases.
  */
 export declare function resetTokenTracking(): void;
 /**

package/dist/utils/tokenTracker.js

@@ -1,6 +1,7 @@
 /**
  * Token and cost tracking for API usage
  */
+import { AsyncLocalStorage } from 'node:async_hooks';
 // Context window sizes per model (in tokens).
 // Keep this table in lockstep with `providers.ts` — entries for models that
 // aren't in the provider catalogue only show up if a user types an id by hand
@@ -10,12 +11,17 @@ const MODEL_CONTEXT_WINDOWS = {
     'glm-5.2': 200_000,
     'glm-5-turbo': 202_752,
     // OpenAI
+    'gpt-5.6-sol': 1_050_000,
+    'gpt-5.6-terra': 1_050_000,
+    'gpt-5.6-luna': 1_050_000,
     'gpt-5.5': 1_200_000,
     'gpt-5.4': 1_050_000,
     'gpt-5.4-mini': 400_000,
     // Anthropic
+    'claude-fable-5': 1_000_000,
     'claude-opus-4-8': 1_000_000,
     'claude-sonnet-4-6': 1_000_000,
+    'claude-sonnet-5': 1_000_000,
     'claude-haiku-4-5-20251001': 200_000,
     // DeepSeek
     'deepseek-v4-pro': 1_000_000,
@@ -23,6 +29,7 @@ const MODEL_CONTEXT_WINDOWS = {
     // Google
     'gemini-3.1-pro-preview': 1_048_576,
     'gemini-3.5-flash': 1_000_000,
+    'gemini-3.1-flash-lite': 1_048_576,
     'gemini-3-flash-preview': 1_000_000,
     // MiniMax
     'MiniMax-M3': 524_288,
@@ -33,6 +40,7 @@ const MODEL_CONTEXT_WINDOWS = {
     'kimi-k2.5': 262_144,
     'kimi-for-coding': 262_144,
     // Grok (xAI)
+    'grok-4.5': 500_000,
     'grok-build-0.1': 256_000,
     'grok-4.3': 1_000_000,
     'grok-code-fast-1': 256_000,
@@ -41,7 +49,7 @@ const MODEL_CONTEXT_WINDOWS = {
     'qwen3-coder-plus': 262_144,
     'qwen3-coder-next': 262_144,
     'qwen3-coder-flash': 262_144,
-    'qwen3-max': 262_144,
+    'qwen3.7-max': 1_000_000,
     'Qwen/Qwen3-Coder-480B-A35B-Instruct': 262_144,
 };
 const DEFAULT_CONTEXT_WINDOW = 128_000;
@@ -63,12 +71,17 @@ const MODEL_PRICING = {
     'glm-5.2': { inputPer1M: 1.00, outputPer1M: 3.20 },
     'glm-5-turbo': { inputPer1M: 1.20, outputPer1M: 4.00 },
     // OpenAI
+    'gpt-5.6-sol': { inputPer1M: 5.00, outputPer1M: 30.00 },
+    'gpt-5.6-terra': { inputPer1M: 2.50, outputPer1M: 15.00 },
+    'gpt-5.6-luna': { inputPer1M: 1.00, outputPer1M: 6.00 },
     'gpt-5.5': { inputPer1M: 5.00, outputPer1M: 30.00 },
     'gpt-5.4': { inputPer1M: 2.50, outputPer1M: 15.00 },
     'gpt-5.4-mini': { inputPer1M: 0.75, outputPer1M: 4.50 },
     // Anthropic
+    'claude-fable-5': { inputPer1M: 10.00, outputPer1M: 50.00 },
     'claude-opus-4-8': { inputPer1M: 5.00, outputPer1M: 25.00 },
     'claude-sonnet-4-6': { inputPer1M: 3.00, outputPer1M: 15.00 },
+    'claude-sonnet-5': { inputPer1M: 3.00, outputPer1M: 15.00 },
     'claude-haiku-4-5-20251001': { inputPer1M: 1.00, outputPer1M: 5.00 },
     // DeepSeek (cache-miss input pricing)
     'deepseek-v4-pro': { inputPer1M: 1.74, outputPer1M: 3.48 },
@@ -76,6 +89,7 @@ const MODEL_PRICING = {
     // Google
     'gemini-3.1-pro-preview': { inputPer1M: 2.00, outputPer1M: 12.00 },
     'gemini-3.5-flash': { inputPer1M: 1.50, outputPer1M: 9.00 },
+    'gemini-3.1-flash-lite': { inputPer1M: 0.25, outputPer1M: 1.50 },
     'gemini-3-flash-preview': { inputPer1M: 0.50, outputPer1M: 3.00 },
     // MiniMax
     'MiniMax-M3': { inputPer1M: 0.60, outputPer1M: 2.40 },
@@ -87,6 +101,7 @@ const MODEL_PRICING = {
     'kimi-k2.5': { inputPer1M: 0.40, outputPer1M: 1.90 },
     'kimi-for-coding': { inputPer1M: 0.60, outputPer1M: 2.50 },
     // Grok (xAI)
+    'grok-4.5': { inputPer1M: 2.00, outputPer1M: 6.00 },
     'grok-build-0.1': { inputPer1M: 1.00, outputPer1M: 2.00 },
     'grok-4.3': { inputPer1M: 1.25, outputPer1M: 2.50 },
     'grok-code-fast-1': { inputPer1M: 0.20, outputPer1M: 1.50 },
@@ -96,15 +111,34 @@ const MODEL_PRICING = {
     'qwen3-coder-plus': { inputPer1M: 0.28, outputPer1M: 1.65 },
     'qwen3-coder-next': { inputPer1M: 0.28, outputPer1M: 1.65 },
     'qwen3-coder-flash': { inputPer1M: 0.10, outputPer1M: 0.50 },
-    'qwen3-max': { inputPer1M: 1.20, outputPer1M: 6.00 },
+    'qwen3.7-max': { inputPer1M: 2.50, outputPer1M: 7.50 },
     // ModelScope free tier — no per-token charge.
     'Qwen/Qwen3-Coder-480B-A35B-Instruct': { inputPer1M: 0, outputPer1M: 0 },
 };
 export function getPricingTable() {
     return Object.entries(MODEL_PRICING).map(([model, p]) => ({ model, ...p }));
 }
-// Session-level accumulator
-const records = [];
+const defaultRecords = [];
+const recordsStore = new AsyncLocalStorage();
+/** The record buffer for the current async flow (a scope's buffer inside
+ *  runWithTokenScope, otherwise the process-wide default). */
+function currentRecords() {
+    return recordsStore.getStore() ?? defaultRecords;
+}
+/** Create a fresh, empty scope buffer (one per ACP session). */
+export function createTokenScope() {
+    return [];
+}
+/**
+ * Run `fn` with `scope` as the active token-record buffer. Every
+ * recordTokenUsage() call made within `fn`'s async flow (including across
+ * awaits) accumulates into `scope`, and reads (getCostBreakdown/…) made in the
+ * same flow see only `scope`. Used by the ACP server to isolate per-session
+ * usage without threading a session id through the deep API layer.
+ */
+export function runWithTokenScope(scope, fn) {
+    return recordsStore.run(scope, fn);
+}
 /**
  * Record token usage from an API response. The optional `actualCostUsd`
  * argument lets aggregator providers (OpenRouter) pass through the
@@ -113,7 +147,7 @@ const records = [];
  * for every OpenRouter-listed model — there are 100+).
  */
 export function recordTokenUsage(usage, model, provider, actualCostUsd) {
-    records.push({
+    currentRecords().push({
         timestamp: Date.now(),
         promptTokens: usage.promptTokens,
         completionTokens: usage.completionTokens,
@@ -130,10 +164,16 @@ export function recordTokenUsage(usage, model, provider, actualCostUsd) {
  */
 export function extractOpenAIUsage(data) {
     if (data?.usage) {
+        // OpenAI-protocol `prompt_tokens` is INCLUSIVE of cached prompt tokens
+        // (DeepSeek/OpenAI report cache hits in prompt_tokens_details.cached_tokens).
+        // Surface them so getCostBreakdown bills cache reads at the discounted
+        // rate instead of the full cache-miss input rate.
+        const cached = data.usage.prompt_tokens_details?.cached_tokens || 0;
         return {
             promptTokens: data.usage.prompt_tokens || 0,
             completionTokens: data.usage.completion_tokens || 0,
             totalTokens: data.usage.total_tokens || 0,
+            cacheReadTokens: cached || undefined,
         };
     }
     return null;
@@ -161,11 +201,16 @@ export function extractAnthropicUsage(data) {
     return null;
 }
 /**
- * Get cost breakdown grouped by provider/model
+ * Get cost breakdown grouped by provider/model.
+ *
+ * `startIndex` lets callers price only the records appended since a marker (see
+ * getRecordCount) — used to report a single run/prompt's delta to cloud
+ * telemetry WITHOUT wiping the session-cumulative store the status bar and
+ * `/cost` read. Defaults to 0 (the whole current scope).
  */
-export function getCostBreakdown() {
+export function getCostBreakdown(startIndex = 0) {
     const grouped = new Map();
-    for (const record of records) {
+    for (const record of currentRecords().slice(startIndex)) {
         const key = `${record.provider}/${record.model}`;
         const existing = grouped.get(key) ?? { provider: record.provider, model: record.model, promptTokens: 0, completionTokens: 0, cacheCreationTokens: 0, cacheReadTokens: 0, estimatedCost: 0 };
         existing.promptTokens += record.promptTokens;
@@ -203,7 +248,7 @@ export function getCacheStats() {
     let cacheCreate = 0;
     let cacheRead = 0;
     let savings = 0;
-    for (const record of records) {
+    for (const record of currentRecords()) {
         cacheCreate += record.cacheCreationTokens ?? 0;
         cacheRead += record.cacheReadTokens ?? 0;
         // Savings = what cache-read tokens would have cost at full input rate,
@@ -227,7 +272,7 @@ export function getSessionStats() {
     let totalTokens = 0;
     let totalCacheCreationTokens = 0;
     let totalCacheReadTokens = 0;
-    for (const record of records) {
+    for (const record of currentRecords()) {
         totalPromptTokens += record.promptTokens;
         totalCompletionTokens += record.completionTokens;
         totalTokens += record.totalTokens;
@@ -239,7 +284,7 @@ export function getSessionStats() {
         totalPromptTokens,
         totalCompletionTokens,
         totalTokens,
-        requestCount: records.length,
+        requestCount: currentRecords().length,
         estimatedCost,
         totalCacheCreationTokens,
         totalCacheReadTokens,
@@ -249,6 +294,7 @@ export function getSessionStats() {
  * Get last request usage
  */
 export function getLastUsage() {
+    const records = currentRecords();
     return records.length > 0 ? records[records.length - 1] : null;
 }
 /**
@@ -262,10 +308,22 @@ export function formatTokenCount(tokens) {
     return (tokens / 1000000).toFixed(2) + 'M';
 }
 /**
- * Reset session tracking
+ * Number of records in the current scope. Capture before a run/prompt and pass
+ * it to getCostBreakdown(startIndex) to price just that run's delta (for cloud
+ * telemetry) without wiping the cumulative store the status bar and `/cost`
+ * read.
+ */
+export function getRecordCount() {
+    return currentRecords().length;
+}
+/**
+ * Reset the current scope's tracking. Production run/prompt paths no longer
+ * call this (they use getRecordCount + getCostBreakdown(startIndex) so the
+ * session-cumulative totals survive); retained for the test suite, which uses
+ * it to isolate the process-wide default buffer between cases.
  */
 export function resetTokenTracking() {
-    records.length = 0;
+    currentRecords().length = 0;
 }
 /**
  * Format a session cost report as a Markdown block. Used by `/cost` in both

package/dist/utils/toolParsing.d.ts

@@ -9,10 +9,21 @@ import { ToolCall } from './tools';
  * Normalize tool name to lowercase with underscores
  */
 export declare function normalizeToolName(name: string): string;
+/**
+ * Extract parameters from truncated/partial JSON for tool calls.
+ * Fallback when JSON.parse fails due to API truncation.
+ */
+declare function extractPartialToolParams(toolName: string, rawArgs: string): Record<string, unknown> | null;
 export declare function parseOpenAIToolCalls(toolCalls: unknown[]): ToolCall[];
 export declare function parseAnthropicToolCalls(content: unknown[]): ToolCall[];
+declare function tryExtractParams(str: string): Record<string, unknown> | null;
+declare function tryParseToolCall(str: string): ToolCall | null;
 /**
  * Parse tool calls from LLM response text.
  * Supports: <tool_call>, <toolcall>, ```tool blocks, inline JSON.
  */
 export declare function parseToolCalls(response: string): ToolCall[];
+export declare const _extractPartialToolParamsForTest: typeof extractPartialToolParams;
+export declare const _tryExtractParamsForTest: typeof tryExtractParams;
+export declare const _tryParseToolCallForTest: typeof tryParseToolCall;
+export {};

package/dist/utils/toolParsing.js

@@ -319,3 +319,9 @@ export function parseToolCalls(response) {
     }
     return toolCalls;
 }
+// Test seams — these helpers are otherwise file-private; export them under
+// a `_forTest` suffix so the parser internals can be exercised directly
+// without going through the full response-parsing pipeline.
+export const _extractPartialToolParamsForTest = extractPartialToolParams;
+export const _tryExtractParamsForTest = tryExtractParams;
+export const _tryParseToolCallForTest = tryParseToolCall;

package/dist/version.d.ts

@@ -1 +1 @@
-export declare const VERSION = "2.13.2";
+export declare const VERSION = "2.15.0";

package/dist/version.js

@@ -1,4 +1,4 @@
 // AUTO-GENERATED by scripts/gen-version.js — do not edit by hand.
 // Baked from package.json at build time so the bun-compiled binary reports
 // the right version (it has no package.json on disk to read at runtime).
-export const VERSION = '2.13.2';
+export const VERSION = '2.15.0';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeep",
-  "version": "2.13.2",
+  "version": "2.15.0",
   "description": "AI-powered coding assistant built for the terminal. Multiple LLM providers, project-aware context, and a seamless development workflow.",
   "type": "module",
   "main": "dist/index.js",
@@ -37,10 +37,10 @@
   },
   "homepage": "https://codeep.dev",
   "dependencies": {
+    "@napi-rs/keyring": "^1.3.0",
     "clipboardy": "^4.0.0",
     "conf": "^13.1.0",
     "js-yaml": "^4.1.0",
-    "keytar": "^7.9.0",
     "open": "^10.0.0"
   },
   "devDependencies": {