@goodandready/dsh-messenger-gateway 0.4.25 → 0.4.27

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/lib/commands.js CHANGED
@@ -24,6 +24,11 @@ export const DEFAULT_TELEGRAM_COMMANDS = [
24
24
  { command: 'fork', description: 'Fork session into new branch' },
25
25
  { command: 'export', description: 'Export history to Markdown' },
26
26
  { command: 'rewind', description: 'Rewind turns: /rewind [N]' },
27
+ { command: 'project', description: 'Bind or show topic workspace: /project' },
28
+ { command: 'memory', description: 'User memory facts: /memory [add <fact>]' },
29
+ { command: 'forget', description: 'Forget user memory: /forget <id|all>' },
30
+ { command: 'outbox', description: 'Outbox queue & DLQ: /outbox [retry|clear]' },
31
+ { command: 'cost', description: 'Session token usage and estimated cost' },
27
32
  { command: 'cwd', description: 'Show working directory: /cwd' },
28
33
  { command: 'files', description: 'Workspace file explorer: /files [dir]' },
29
34
  { command: 'get', description: 'Download file to messenger: /get <path>' },
@@ -104,6 +109,11 @@ export const HELP_TEXT = [
104
109
  '',
105
110
  '🛠️ <b>Tools, Files & Cron:</b>',
106
111
  '• /skills / /tools — list active tools & skills',
112
+ '• /project — topic workspace binding (/project bind <path>)',
113
+ '• /memory — show remembered facts about you (/memory add <fact>)',
114
+ '• /forget <id|all> — delete stored user facts',
115
+ '• /outbox — transactional queue & DLQ (/outbox [retry|clear])',
116
+ '• /cost — session token usage, breakdown & estimated cost',
107
117
  '• /cwd — show current working directory',
108
118
  '• /files [dir] — workspace file explorer',
109
119
  '• /get <path> — download file from workspace',
package/lib/config.js CHANGED
@@ -58,6 +58,8 @@ export const PluginConfig = z.object({
58
58
  .description('mirror: TTS when inbound was voice; always/off override (per-user /voice wins)'),
59
59
  quickActions: z.boolean().default(false).description('Show quick actions keyboard in Telegram (/new, /stop, /voice, /status)'),
60
60
  artifactPreviews: z.boolean().default(true).description('Render diagrams and formatted tables as previews'),
61
+ costFooter: z.boolean().default(false).description('Append compact execution & cost footer to bot replies (e.g. ⏱️ 2.4s · 1.8k tokens · deepseek-chat)'),
62
+ budgetAlertThresholdUsd: z.number().default(0).description('Session budget warning threshold in USD (0 = disabled)'),
61
63
  notifyBridge: z.object({
62
64
  enabled: z.boolean().default(false),
63
65
  events: z.array(z.string()).default(['task_done', 'error']),
@@ -103,6 +105,15 @@ export const PluginConfig = z.object({
103
105
  maxChars: z.number().default(4000),
104
106
  voiceSummary: z.boolean().default(false).description('Speak concise TL;DR summary while full text is sent to chat'),
105
107
  }).volatile(),
108
+ memory: z.object({
109
+ enabled: z.boolean().default(true),
110
+ baseUrl: z.string().default('').description('Base URL for dsh-memory-brain service (default http://127.0.0.1:28080)'),
111
+ tokenEnv: z.string().default('MEMORY_BRAIN_TOKEN').description('Environment variable name containing the bearer token'),
112
+ token: z.string().role('secret').default('').description('Explicit bearer token if not using tokenEnv'),
113
+ autoInject: z.boolean().default(true).description('Automatically retrieve and inject user memories into agent turn context'),
114
+ budget: z.number().default(2000).description('Character budget for injected memories'),
115
+ timeoutMs: z.number().default(3500).description('HTTP request timeout in ms for memory-brain calls'),
116
+ }).default({ enabled: true }).volatile(),
106
117
  agent: z.object({
107
118
  provider: z.string().default(''),
108
119
  model: z.string().default(''),
@@ -0,0 +1,228 @@
1
+ /**
2
+ * Token usage and cost tracker in messenger sessions (dsh-cost-meter bridge, #156).
3
+ *
4
+ * Provides:
5
+ * - Accurate token extraction from turn events and DSH session projections.
6
+ * - Model cost estimation using standard rate cards and dsh-cost-meter tariffs.
7
+ * - Compact response footer (e.g. ⏱️ 2.4s · 1.8k tokens (~$0.004) · deepseek-chat).
8
+ * - Detailed `/cost` command with session breakdown and daily budget tracking.
9
+ * - Budget threshold alerts when nearing session limits.
10
+ */
11
+
12
+ import { escapeHtml } from './telegram-format.js'
13
+
14
+ export const DEFAULT_MODEL_RATES = {
15
+ 'deepseek-chat': { input: 0.44, output: 1.32, cacheRead: 0.014, cacheWrite: 0.44 },
16
+ 'deepseek-v3': { input: 0.44, output: 1.32, cacheRead: 0.014, cacheWrite: 0.44 },
17
+ 'deepseek-reasoner': { input: 1.32, output: 3.96, cacheRead: 0.044, cacheWrite: 1.32 },
18
+ 'deepseek-r1': { input: 1.32, output: 3.96, cacheRead: 0.044, cacheWrite: 1.32 },
19
+ 'gpt-4o': { input: 2.50, output: 10.00, cacheRead: 1.25, cacheWrite: 2.50 },
20
+ 'gpt-4o-mini': { input: 0.15, output: 0.60, cacheRead: 0.075, cacheWrite: 0.15 },
21
+ 'claude-3-5-sonnet': { input: 3.00, output: 15.00, cacheRead: 0.30, cacheWrite: 3.75 },
22
+ 'fallback': { input: 0.50, output: 1.50, cacheRead: 0.05, cacheWrite: 0.50 },
23
+ }
24
+
25
+ export function resolveModelRates(modelId) {
26
+ if (!modelId || typeof modelId !== 'string') return DEFAULT_MODEL_RATES.fallback
27
+ const clean = modelId.toLowerCase().split('/').pop()
28
+ for (const [key, rates] of Object.entries(DEFAULT_MODEL_RATES)) {
29
+ if (clean.includes(key)) return rates
30
+ }
31
+ return DEFAULT_MODEL_RATES.fallback
32
+ }
33
+
34
+ export function normalizeUsage(rawUsage) {
35
+ if (!rawUsage || typeof rawUsage !== 'object') {
36
+ return {
37
+ inputTokens: 0,
38
+ outputTokens: 0,
39
+ cacheReadTokens: 0,
40
+ cacheWriteTokens: 0,
41
+ totalTokens: 0,
42
+ }
43
+ }
44
+
45
+ const inputTokens = Number(rawUsage.inputTokens || rawUsage.prompt_tokens || rawUsage.promptTokens || rawUsage.input || 0)
46
+ const outputTokens = Number(rawUsage.outputTokens || rawUsage.completion_tokens || rawUsage.completionTokens || rawUsage.output || 0)
47
+ const cacheReadTokens = Number(rawUsage.cacheReadTokens || rawUsage.cache_read_tokens || rawUsage.cached_tokens || 0)
48
+ const cacheWriteTokens = Number(rawUsage.cacheWriteTokens || rawUsage.cache_creation_input_tokens || 0)
49
+ const totalTokens = Number(rawUsage.totalTokens || rawUsage.total_tokens || (inputTokens + outputTokens + cacheReadTokens + cacheWriteTokens))
50
+
51
+ return {
52
+ inputTokens,
53
+ outputTokens,
54
+ cacheReadTokens,
55
+ cacheWriteTokens,
56
+ totalTokens,
57
+ }
58
+ }
59
+
60
+ export function calculateEstimatedCostUsd(usage, modelId) {
61
+ const norm = normalizeUsage(usage)
62
+ const rates = resolveModelRates(modelId)
63
+
64
+ const inputCost = (norm.inputTokens * rates.input) / 1_000_000
65
+ const outputCost = (norm.outputTokens * rates.output) / 1_000_000
66
+ const cacheReadCost = (norm.cacheReadTokens * rates.cacheRead) / 1_000_000
67
+ const cacheWriteCost = (norm.cacheWriteTokens * rates.cacheWrite) / 1_000_000
68
+
69
+ return inputCost + outputCost + cacheReadCost + cacheWriteCost
70
+ }
71
+
72
+ export function formatTokenCount(n) {
73
+ const count = Number(n) || 0
74
+ if (count < 1000) return String(count)
75
+ const k = count / 1000
76
+ return (k >= 10 ? k.toFixed(0) : k.toFixed(1)) + 'k'
77
+ }
78
+
79
+ export function formatDurationSeconds(ms) {
80
+ const sec = Math.max(0.1, (Number(ms) || 0) / 1000)
81
+ return sec >= 10 ? `${sec.toFixed(0)}s` : `${sec.toFixed(1)}s`
82
+ }
83
+
84
+ export function formatCostUsd(usd) {
85
+ const val = Number(usd) || 0
86
+ if (val <= 0) return '$0.00'
87
+ if (val < 0.001) return '<$0.001'
88
+ return `~$${val.toFixed(4)}`
89
+ }
90
+
91
+ export function formatCostFooter({ durationMs, usage, model }) {
92
+ const norm = normalizeUsage(usage)
93
+ if (!norm.totalTokens && !durationMs) return ''
94
+
95
+ const durStr = formatDurationSeconds(durationMs)
96
+ const tokenStr = formatTokenCount(norm.totalTokens)
97
+ const cost = calculateEstimatedCostUsd(norm, model)
98
+ const costStr = formatCostUsd(cost)
99
+ const modelName = (model || 'model').split('/').pop()
100
+
101
+ return `\n\n<i>⏱️ ${durStr} · ${tokenStr} tokens (${costStr}) · ${escapeHtml(modelName)}</i>`
102
+ }
103
+
104
+ export function extractSessionStats(chat) {
105
+ let promptTokens = 0
106
+ let completionTokens = 0
107
+ let cachedTokens = 0
108
+ let totalTokens = 0
109
+ let turnsCount = 0
110
+
111
+ // 1. Try DSH session projections if available
112
+ const session = chat?.agent?.session
113
+ let projection = null
114
+ try {
115
+ if (typeof session?.projection === 'function') {
116
+ projection = session.projection('costByModel')
117
+ }
118
+ } catch {
119
+ /* projection not registered or session closed */
120
+ }
121
+
122
+ if (projection?.models && Array.isArray(projection.models)) {
123
+ for (const m of projection.models) {
124
+ if (m.slots && typeof m.slots === 'object') {
125
+ for (const slot of Object.values(m.slots)) {
126
+ promptTokens += Number(slot.uncachedInputTokens || 0)
127
+ completionTokens += Number(slot.outputTokens || 0)
128
+ cachedTokens += Number(slot.cacheReadTokens || 0)
129
+ }
130
+ }
131
+ }
132
+ if (Array.isArray(projection.turns)) {
133
+ turnsCount = projection.turns.length
134
+ }
135
+ }
136
+
137
+ // 2. Accumulate chat turn metrics if projection empty
138
+ if (!promptTokens && !completionTokens && chat?.turnStats) {
139
+ promptTokens = chat.turnStats.inputTokens || 0
140
+ completionTokens = chat.turnStats.outputTokens || 0
141
+ cachedTokens = chat.turnStats.cacheReadTokens || 0
142
+ turnsCount = chat.turnStats.turnsCount || 0
143
+ }
144
+
145
+ // 3. Fallback to last turn usage if available
146
+ if (!promptTokens && !completionTokens && chat?.lastTurnUsage) {
147
+ const lastNorm = normalizeUsage(chat.lastTurnUsage)
148
+ promptTokens = lastNorm.inputTokens
149
+ completionTokens = lastNorm.outputTokens
150
+ cachedTokens = lastNorm.cacheReadTokens
151
+ turnsCount = 1
152
+ }
153
+
154
+ totalTokens = promptTokens + completionTokens + cachedTokens
155
+
156
+ return {
157
+ promptTokens,
158
+ completionTokens,
159
+ cachedTokens,
160
+ totalTokens,
161
+ turnsCount,
162
+ }
163
+ }
164
+
165
+ export function formatSessionCostReport({ chat, config, model }) {
166
+ const stats = extractSessionStats(chat)
167
+ const activeModel = model || config?.agent?.model || 'deepseek-chat'
168
+ const costUsd = calculateEstimatedCostUsd({
169
+ inputTokens: stats.promptTokens,
170
+ outputTokens: stats.completionTokens,
171
+ cacheReadTokens: stats.cachedTokens,
172
+ }, activeModel)
173
+
174
+ const budgetThreshold = Number(config?.telegram?.budgetAlertThresholdUsd || 0)
175
+
176
+ const lines = [
177
+ '📊 <b>Session Token & Cost Meter:</b>',
178
+ `• Active Model: <code>${escapeHtml(activeModel)}</code>`,
179
+ `• Total Tokens: <b>${formatTokenCount(stats.totalTokens)}</b> (${stats.totalTokens.toLocaleString()})`,
180
+ ` ├─ Prompt (input): ${formatTokenCount(stats.promptTokens)}`,
181
+ ` ├─ Completion (output): ${formatTokenCount(stats.completionTokens)}`,
182
+ ` └─ Cached read: ${formatTokenCount(stats.cachedTokens)}`,
183
+ `• Completed Turns: <b>${stats.turnsCount}</b>`,
184
+ `• Estimated Cost: <b>${formatCostUsd(costUsd)}</b> USD`,
185
+ ]
186
+
187
+ if (budgetThreshold > 0) {
188
+ const remaining = Math.max(0, budgetThreshold - costUsd)
189
+ const pct = Math.min(100, (costUsd / budgetThreshold) * 100).toFixed(1)
190
+ lines.push(
191
+ '',
192
+ '🎯 <b>Budget Tracking:</b>',
193
+ `• Budget Threshold: <b>$${budgetThreshold.toFixed(2)}</b>`,
194
+ `• Consumed: <b>${pct}%</b>`,
195
+ `• Remaining: <b>$${remaining.toFixed(4)}</b>`,
196
+ )
197
+ if (costUsd >= budgetThreshold) {
198
+ lines.push('⚠️ <b>Budget threshold reached!</b> Consider using `/new` to reset context.')
199
+ }
200
+ }
201
+
202
+ lines.push('', '<i>Run <code>/cost</code> anytime to check current session spend.</i>')
203
+ return lines.join('\n')
204
+ }
205
+
206
+ export function checkBudgetThreshold({ currentSpendUsd, thresholdUsd }) {
207
+ const threshold = Number(thresholdUsd) || 0
208
+ if (threshold <= 0) return null
209
+ const spend = Number(currentSpendUsd) || 0
210
+
211
+ if (spend >= threshold) {
212
+ return {
213
+ alert: true,
214
+ level: 'limit',
215
+ message: `Session cost ($${spend.toFixed(4)}) has exceeded the configured budget ($${threshold.toFixed(2)}).`,
216
+ }
217
+ }
218
+
219
+ if (spend >= threshold * 0.8) {
220
+ return {
221
+ alert: true,
222
+ level: 'warning',
223
+ message: `Session cost ($${spend.toFixed(4)}) has reached 80% of budget limit ($${threshold.toFixed(2)}).`,
224
+ }
225
+ }
226
+
227
+ return null
228
+ }
@@ -1,4 +1,5 @@
1
- import { resolve, dirname } from 'node:path'
1
+ import { resolve, dirname, basename } from 'node:path'
2
+ import { realpathSync } from 'node:fs'
2
3
  import { fileURLToPath } from 'node:url'
3
4
  import { ensureContentArray } from './content-guard.js'
4
5
  import { t } from './locales/index.js'
@@ -12,8 +13,10 @@ import { listModelCatalog, buildProvidersKeyboard } from './models.js'
12
13
  import { normalizeHomeName, upsertHome, listHomes } from './homes.js'
13
14
  import { handleSetAlertCommand, handleAlertCommand } from './alerts.js'
14
15
  import { HELP_TEXT } from './commands.js'
15
- import { resolveEffectiveCwd } from './workspace.js'
16
+ import { resolveEffectiveCwd, getCwdWarning, inspectGitWorkspace, registerSessionInWorkspace } from './workspace.js'
16
17
  import { escapeHtml } from './telegram-format.js'
18
+ import { rememberUserFact, listUserMemories, forgetUserMemory } from './memory-brain.js'
19
+ import { formatSessionCostReport } from './cost-tracker.js'
17
20
 
18
21
 
19
22
  export function releaseChatTurn(chat) {
@@ -252,14 +255,208 @@ export async function handleGatewayCommand(gw, key, text, input) {
252
255
  }
253
256
  }
254
257
 
258
+ if (cmd === '/project') {
259
+ const sub = parts[1]?.toLowerCase()
260
+ if (sub === 'bind') {
261
+ const targetPath = parts[2]
262
+ if (!targetPath) {
263
+ return reply('Usage: <code>/project bind &lt;path&gt; [role]</code>\nExample: <code>/project bind /home/vadim/my-app coder</code>')
264
+ }
265
+ const warning = getCwdWarning(targetPath)
266
+ if (warning) {
267
+ return reply(`❌ <b>Invalid workspace directory:</b>\n${escapeHtml(warning)}`)
268
+ }
269
+ let realCwd = targetPath
270
+ try { realCwd = realpathSync(resolve(targetPath)) } catch { realCwd = resolve(targetPath) }
271
+ const gitInfo = inspectGitWorkspace(realCwd)
272
+ const roleArg = parts[3]?.toLowerCase()
273
+ const validRole = roleArg && getPersona(roleArg) ? roleArg : null
274
+
275
+ gw.topicWorkspaces?.set(chatId, threadId, {
276
+ path: realCwd,
277
+ name: basename(realCwd),
278
+ role: validRole,
279
+ worktree: gitInfo.isWorktree,
280
+ })
281
+
282
+ if (validRole) {
283
+ gw.personas?.set(chatId, validRole, threadId)
284
+ }
285
+
286
+ const chat = gw.chats.get(key)
287
+ if (chat?.agent?.session?.id) {
288
+ registerSessionInWorkspace(gw.ctx, chat.agent.session.id, realCwd, gw.ctx?.logger).catch(() => {})
289
+ }
290
+
291
+ const lines = [
292
+ `✅ <b>Workspace bound to topic:</b>`,
293
+ `📁 Directory: <code>${escapeHtml(realCwd)}</code>`,
294
+ ]
295
+ if (gitInfo.isWorktree) lines.push(`🌿 Git Worktree: <code>detected</code>`)
296
+ else if (gitInfo.isGit) lines.push(`📦 Git Repository: <code>main repo</code>`)
297
+ if (validRole) lines.push(`🎭 Persona role: <b>${validRole}</b>`)
298
+ return reply(lines.join('\n'))
299
+ }
300
+
301
+ if (sub === 'unbind' || sub === 'reset' || sub === 'clear') {
302
+ const removed = gw.topicWorkspaces?.delete(chatId, threadId)
303
+ if (removed) {
304
+ return reply('✅ <b>Workspace unbind:</b> topic reverted to default workspace.')
305
+ }
306
+ return reply('ℹ️ No custom workspace was bound to this topic.')
307
+ }
308
+
309
+ if (sub === 'list') {
310
+ const list = gw.topicWorkspaces?.list(chatId) || []
311
+ if (!list.length) {
312
+ return reply('ℹ️ No topics in this chat have bound workspaces.')
313
+ }
314
+ const lines = ['📁 <b>Bound Topic Workspaces:</b>\n']
315
+ for (const item of list) {
316
+ const tLabel = item.threadId ? `Topic #${item.threadId}` : 'Main Chat'
317
+ const roleLabel = item.role ? ` (role: ${item.role})` : ''
318
+ lines.push(`• <b>${tLabel}</b>: <code>${escapeHtml(item.path)}</code>${roleLabel}`)
319
+ }
320
+ return reply(lines.join('\n'))
321
+ }
322
+
323
+ const bound = gw.topicWorkspaces?.get(chatId, threadId)
324
+ const effective = gw.resolveEffectiveCwdFor ? gw.resolveEffectiveCwdFor(chatId, threadId) : resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger)
325
+ const lines = [
326
+ `📁 <b>Project Workspace Status:</b>`,
327
+ ]
328
+ if (bound) {
329
+ lines.push(`• Bound path: <code>${escapeHtml(bound.path)}</code>`)
330
+ if (bound.worktree) lines.push(`• Mode: 🌿 Git Worktree`)
331
+ if (bound.role) lines.push(`• Bound role: <b>${bound.role}</b>`)
332
+ lines.push(`• Bound since: <code>${bound.boundAt}</code>`)
333
+ } else {
334
+ lines.push(`• Current workspace: <code>${escapeHtml(effective)}</code> (default)`)
335
+ lines.push(`• Status: <i>Not bound to dedicated project</i>`)
336
+ }
337
+ lines.push('\n<b>Commands:</b>')
338
+ lines.push('• <code>/project bind &lt;path&gt; [role]</code> — bind topic to directory or worktree')
339
+ lines.push('• <code>/project unbind</code> — reset topic to default workspace')
340
+ lines.push('• <code>/project list</code> — view all topic bindings')
341
+ return reply(lines.join('\n'))
342
+ }
343
+
344
+ if (cmd === '/memory') {
345
+ const sub = parts[1]?.toLowerCase()
346
+ if (sub === 'add') {
347
+ const factText = parts.slice(2).join(' ').trim()
348
+ if (!factText) {
349
+ return reply('Usage: <code>/memory add &lt;fact&gt;</code>\nExample: <code>/memory add I prefer concise answers in Python</code>')
350
+ }
351
+ const res = await rememberUserFact({
352
+ config: gw.config?.memory,
353
+ userId,
354
+ content: factText,
355
+ })
356
+ if (!res.ok) {
357
+ return reply(`❌ Failed to store memory: ${escapeHtml(res.error || 'unknown error')}`)
358
+ }
359
+ return reply(`🧠 <b>Memory remembered:</b>\n<i>${escapeHtml(factText)}</i>`)
360
+ }
361
+
362
+ const listRes = await listUserMemories({
363
+ config: gw.config?.memory,
364
+ userId,
365
+ limit: 25,
366
+ })
367
+ if (!listRes.ok) {
368
+ return reply(`⚠️ Memory service unavailable: ${escapeHtml(listRes.error || 'could not connect')}`)
369
+ }
370
+ if (!listRes.items || !listRes.items.length) {
371
+ return reply(`🧠 <b>User Memory (ID: <code>${escapeHtml(String(userId))}</code>):</b>\n<i>No stored facts found.</i>\n\nAdd one: <code>/memory add &lt;fact&gt;</code>`)
372
+ }
373
+ const lines = [
374
+ `🧠 <b>User Memory Profile (ID: <code>${escapeHtml(String(userId))}</code>):</b>`,
375
+ '',
376
+ ...listRes.items.map((it, idx) => `${idx + 1}. [<code>${escapeHtml(it.id.slice(0, 8))}</code>] ${escapeHtml(it.content)}`),
377
+ '',
378
+ '• Add fact: <code>/memory add &lt;fact&gt;</code>',
379
+ '• Forget fact: <code>/forget &lt;id&gt;</code>',
380
+ '• Clear all: <code>/forget all</code>',
381
+ ]
382
+ return reply(lines.join('\n'))
383
+ }
384
+
385
+ if (cmd === '/forget') {
386
+ const targetId = parts[1]?.trim()
387
+ if (!targetId) {
388
+ return reply('Usage: <code>/forget &lt;id|all&gt;</code>\nExample: <code>/forget 95b82eb8</code> or <code>/forget all</code>\nRun <code>/memory</code> to view memory IDs.')
389
+ }
390
+ const forgetRes = await forgetUserMemory({
391
+ config: gw.config?.memory,
392
+ userId,
393
+ memoryId: targetId,
394
+ })
395
+ if (!forgetRes.ok) {
396
+ return reply(`❌ Failed to forget: ${escapeHtml(forgetRes.error || 'unknown error')}`)
397
+ }
398
+ if (targetId.toLowerCase() === 'all') {
399
+ return reply(`🗑️ Forgotten ${forgetRes.count || 0} user memor${forgetRes.count === 1 ? 'y' : 'ies'}.`)
400
+ }
401
+ return reply(`🗑️ Forgotten memory <code>${escapeHtml(targetId)}</code>.`)
402
+ }
403
+
404
+ if (cmd === '/outbox') {
405
+ const sub = parts[1]?.toLowerCase()
406
+ if (!gw.outbox) {
407
+ return reply('ℹ️ Outbox queue is not initialized.')
408
+ }
409
+
410
+ if (sub === 'retry') {
411
+ const count = await gw.outbox.retryDlq()
412
+ return reply(`🔄 Moved <b>${count}</b> DLQ items back to outbox queue for retry.`)
413
+ }
414
+
415
+ if (sub === 'clear') {
416
+ const count = await gw.outbox.clearDlq()
417
+ return reply(`🗑️ Cleared <b>${count}</b> items from Dead Letter Queue.`)
418
+ }
419
+
420
+ const st = gw.outbox.status()
421
+ const lines = [
422
+ '📬 <b>Outbox Transactional Queue:</b>',
423
+ `• Pending messages: <b>${st.pendingCount}</b>`,
424
+ `• Dead Letter Queue (DLQ): <b>${st.dlqCount}</b>`,
425
+ ]
426
+ if (st.dlqCount > 0) {
427
+ lines.push('', '<b>Recent DLQ Failures:</b>')
428
+ const sample = st.dlq.slice(0, 5)
429
+ for (const item of sample) {
430
+ lines.push(`- [<code>${escapeHtml(item.id.slice(0, 8))}</code>] to ${escapeHtml(String(item.target?.chatId))}: ${escapeHtml(item.error || 'unknown error')}`)
431
+ }
432
+ lines.push('', 'Commands:', '• Retry all: <code>/outbox retry</code>', '• Clear DLQ: <code>/outbox clear</code>')
433
+ } else {
434
+ lines.push('', '<i>All outbound messages successfully delivered.</i>')
435
+ }
436
+ return reply(lines.join('\n'))
437
+ }
438
+
439
+ if (cmd === '/cost') {
440
+ const chat = gw.chats.get(key)
441
+ const report = formatSessionCostReport({
442
+ chat,
443
+ config: gw.config,
444
+ model: gw.config.agent?.model || 'deepseek-chat',
445
+ })
446
+ return reply(report)
447
+ }
448
+
255
449
  if (cmd === '/cwd') {
256
450
  const chat = gw.chats.get(key)
257
- const sessionCwd = chat?.agent?.meta?.cwd || chat?.agent?.session?.meta?.cwd || resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger)
451
+ const sessionCwd = chat?.agent?.meta?.cwd || chat?.agent?.session?.meta?.cwd || (gw.resolveEffectiveCwdFor ? gw.resolveEffectiveCwdFor(chatId, threadId) : resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger))
452
+ const bound = gw.topicWorkspaces?.get(chatId, threadId)
258
453
  const configuredCwd = gw.config.agent?.cwd || ''
259
454
  const lines = [
260
455
  `📁 <b>Working Directory:</b> <code>${escapeHtml(sessionCwd)}</code>`,
261
456
  ]
262
- if (configuredCwd) {
457
+ if (bound) {
458
+ lines.push(`🌿 Topic binding: <code>${escapeHtml(bound.path)}</code>${bound.worktree ? ' (worktree)' : ''}`)
459
+ } else if (configuredCwd) {
263
460
  lines.push(`⚙️ Configured: <code>${escapeHtml(configuredCwd)}</code>`)
264
461
  } else {
265
462
  lines.push(`ℹ️ Default: host launch directory`)
@@ -269,14 +466,14 @@ export async function handleGatewayCommand(gw, key, text, input) {
269
466
 
270
467
  if (cmd === '/files') {
271
468
  const subPath = parts.slice(1).join(' ').trim() || '.'
272
- const agentCwd = resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger)
469
+ const agentCwd = (gw.resolveEffectiveCwdFor ? gw.resolveEffectiveCwdFor(chatId, threadId) : resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger))
273
470
  const res = await handleFilesCommand(agentCwd, subPath)
274
471
  return reply(res.text)
275
472
  }
276
473
 
277
474
  if (cmd === '/get') {
278
475
  const targetRel = parts.slice(1).join(' ').trim()
279
- const agentCwd = resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger)
476
+ const agentCwd = (gw.resolveEffectiveCwdFor ? gw.resolveEffectiveCwdFor(chatId, threadId) : resolveEffectiveCwd(gw.config.agent?.cwd, gw.ctx?.logger))
280
477
  const maxDocBytes = Number(gw.config.media?.maxDocBytes) || 50 * 1024 * 1024
281
478
  const res = await handleGetCommand(agentCwd, targetRel, maxDocBytes)
282
479
  return reply(res.files ? { text: res.text, files: res.files } : res.text)
@@ -15,6 +15,8 @@ import {
15
15
  buildStreamPreview, formatProgressLine, createEditScheduler, startTypingHeartbeat, parseReasoningStream
16
16
  } from './stream.js'
17
17
  import { escapeHtml } from './telegram-format.js'
18
+ import { formatCostFooter, normalizeUsage, extractSessionStats, calculateEstimatedCostUsd, checkBudgetThreshold } from './cost-tracker.js'
19
+ import { recallUserMemories, recordTurnMemory } from './memory-brain.js'
18
20
 
19
21
  const PLUGIN = 'dsh-messenger-gateway'
20
22
 
@@ -135,9 +137,23 @@ export async function answerApproval(gw, chatKeyValue, req, next) {
135
137
  }
136
138
 
137
139
  export async function buildUserContent(gw, input, signal) {
138
- const { text, attachments = [], replyText, steer, personaOverride } = input
140
+ const { text, attachments = [], replyText, steer, personaOverride, userId } = input
139
141
  const parts = []
140
142
  parts.push(String(gw.config.agent?.instructionPrefix || MESSENGER_RELAY_INSTRUCTION))
143
+ if (gw.config?.memory?.enabled !== false && gw.config?.memory?.autoInject !== false && userId) {
144
+ try {
145
+ const recalled = await recallUserMemories({
146
+ config: gw.config.memory,
147
+ userId,
148
+ query: text || replyText || '',
149
+ budget: gw.config.memory?.budget,
150
+ signal,
151
+ })
152
+ if (recalled) parts.push(recalled)
153
+ } catch (memErr) {
154
+ gw.ctx?.logger?.debug?.(`[memory-brain] recall error: ${memErr?.message || memErr}`)
155
+ }
156
+ }
141
157
  const activePersonaId = personaOverride || gw.personas.getPersonaForChat(input.chatId, input.threadId)
142
158
  const activePersona = getPersona(activePersonaId)
143
159
  if (activePersona?.instruction) {
@@ -200,6 +216,7 @@ export async function buildUserContent(gw, input, signal) {
200
216
 
201
217
  export async function runGatewayTurn(gw, chat, input, signal) {
202
218
  const { reply, typing, startStream, startProgress, react, inboundWasVoice, userId } = input
219
+ const turnStartTime = Date.now()
203
220
  chat.turnActive = true
204
221
  const sessionId = chat.agent.session.id
205
222
  const tg = gw.tg()
@@ -315,6 +332,23 @@ export async function runGatewayTurn(gw, chat, input, signal) {
315
332
  else await reply(noResp)
316
333
  return
317
334
  }
335
+ if (collector.usage) {
336
+ chat.lastTurnUsage = collector.usage
337
+ if (!chat.turnStats) chat.turnStats = { inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, turnsCount: 0 }
338
+ chat.turnStats.turnsCount++
339
+ const norm = normalizeUsage(collector.usage)
340
+ chat.turnStats.inputTokens += norm.inputTokens
341
+ chat.turnStats.outputTokens += norm.outputTokens
342
+ chat.turnStats.cacheReadTokens += norm.cacheReadTokens
343
+ }
344
+ if (tg.costFooter && answer) {
345
+ const footer = formatCostFooter({
346
+ durationMs: Date.now() - turnStartTime,
347
+ usage: collector.usage,
348
+ model: gw.config?.agent?.model,
349
+ })
350
+ if (footer) answer += footer
351
+ }
318
352
  const maxLen = Number(gw.config.agent?.maxMessageLength) || 4000
319
353
  const chunks = answer ? splitText(answer, maxLen) : ['']
320
354
  if (stream) {
@@ -348,6 +382,34 @@ export async function runGatewayTurn(gw, chat, input, signal) {
348
382
  }
349
383
  }
350
384
  }
385
+ const budgetThreshold = Number(tg.budgetAlertThresholdUsd || 0)
386
+ if (budgetThreshold > 0) {
387
+ const stats = extractSessionStats(chat)
388
+ const cost = calculateEstimatedCostUsd({
389
+ inputTokens: stats.promptTokens,
390
+ outputTokens: stats.completionTokens,
391
+ cacheReadTokens: stats.cachedTokens,
392
+ }, gw.config?.agent?.model)
393
+ const warning = checkBudgetThreshold({ currentSpendUsd: cost, thresholdUsd: budgetThreshold })
394
+ if (warning && !chat.budgetWarned) {
395
+ chat.budgetWarned = true
396
+ gw.sendAlert('status', {
397
+ title: 'Budget Alert',
398
+ details: `${warning.message} (Chat: ${input.chatId})`,
399
+ }).catch(() => {})
400
+ }
401
+ }
402
+ if (gw.config?.memory?.enabled !== false && userId && answer && input.text) {
403
+ recordTurnMemory({
404
+ config: gw.config.memory,
405
+ userId,
406
+ userText: input.text,
407
+ assistantText: answer,
408
+ logger: gw.ctx?.logger,
409
+ }).catch?.((recErr) => {
410
+ gw.ctx?.logger?.debug?.(`[memory-brain] recordTurnMemory error: ${recErr?.message || recErr}`)
411
+ })
412
+ }
351
413
  } catch (err) {
352
414
  if (!signal?.aborted) {
353
415
  gw.sendAlert('error', {