feat: add session export with full and compressed modes (#507)

Add export functionality that allows users to download session data
as JSON or plain text, with optional LLM-based context compression
for long conversations. Includes UI controls in chat panel, session
list, and history view, plus i18n strings for all 8 locales.

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

packages/server/src/lib/context-compressor/export-compressor.ts

@@ -0,0 +1,149 @@
+/**
+ * Export Compressor
+ *
+ * Compresses session context for export purposes.
+ * Reuses the LLM summarization logic from ChatContextCompressor
+ * but does NOT read or write compression snapshots.
+ * Always forces LLM compression regardless of token count.
+ * No tail reservation — all messages are compressed.
+ */
+
+import { logger } from '../../services/logger'
+import {
+  type ChatMessage,
+  type CompressionConfig,
+  type CompressedResult,
+  DEFAULT_COMPRESSION_CONFIG,
+  countTokens,
+  serializeForSummary,
+  buildFullPrompt,
+  buildIncrementalPrompt,
+  buildConversationHistory,
+  callSummarizer,
+} from './index'
+import { getCompressionSnapshot } from '../../db/hermes/compression-snapshot'
+
+export class ExportCompressor {
+  private config: CompressionConfig
+
+  constructor(opts?: { config?: Partial<CompressionConfig> }) {
+    this.config = { ...DEFAULT_COMPRESSION_CONFIG, ...opts?.config }
+  }
+
+  async compress(
+    messages: ChatMessage[],
+    upstream: string,
+    apiKey: string | undefined,
+    sessionId?: string,
+    profile?: string,
+  ): Promise<CompressedResult> {
+    const total = messages.length
+
+    const meta: CompressedResult['meta'] = {
+      totalMessages: total,
+      compressed: false,
+      llmCompressed: false,
+      summaryTokenEstimate: 0,
+      verbatimCount: 0,
+      compressedStartIndex: -1,
+    }
+
+    // Read snapshot for incremental context, but never write
+    const snapshot = sessionId ? getCompressionSnapshot(sessionId) : null
+
+    if (snapshot) {
+      logger.info(
+        '[export-compressor] session=%s: incremental compress with existing snapshot at index %d',
+        sessionId, snapshot.lastMessageIndex,
+      )
+      return this.incrementalCompress(
+        messages, snapshot, upstream, apiKey, meta, profile,
+      )
+    }
+
+    logger.info(
+      '[export-compressor] session=%s: full compress %d messages',
+      sessionId, total,
+    )
+    return this.fullCompress(messages, upstream, apiKey, meta, profile)
+  }
+
+  private async incrementalCompress(
+    messages: ChatMessage[],
+    snapshot: { summary: string; lastMessageIndex: number },
+    upstream: string,
+    apiKey: string | undefined,
+    meta: CompressedResult['meta'],
+    profile?: string,
+  ): Promise<CompressedResult> {
+    const { summary: previousSummary, lastMessageIndex } = snapshot
+    const newMessages = messages.slice(lastMessageIndex + 1)
+
+    let summary: string | null = null
+    try {
+      const contentToSummarize = serializeForSummary(newMessages)
+      const prompt = buildIncrementalPrompt(previousSummary, contentToSummarize, this.config.summaryBudget)
+      const history = buildConversationHistory(newMessages)
+
+      const t0 = Date.now()
+      summary = await callSummarizer(upstream, apiKey, prompt, history, this.config.summarizationTimeoutMs, previousSummary, profile)
+      logger.info('[export-compressor] incremental-llm done in %dms, %d chars', Date.now() - t0, summary!.length)
+    } catch (err: any) {
+      logger.warn('[export-compressor] incremental-llm failed: %s — reusing previous summary', err.message)
+      summary = previousSummary
+    }
+
+    const summaryText = summary || previousSummary
+
+    return {
+      messages: [{ role: 'user', content: summaryText }],
+      meta: {
+        ...meta,
+        compressed: true,
+        llmCompressed: true,
+        summaryTokenEstimate: countTokens(summaryText),
+        verbatimCount: 0,
+      },
+    }
+  }
+
+  private async fullCompress(
+    messages: ChatMessage[],
+    upstream: string,
+    apiKey: string | undefined,
+    meta: CompressedResult['meta'],
+    profile?: string,
+  ): Promise<CompressedResult> {
+    if (messages.length === 0) {
+      return { messages: [], meta }
+    }
+
+    let summary: string | null = null
+    try {
+      const contentToSummarize = serializeForSummary(messages)
+      const prompt = buildFullPrompt(contentToSummarize, this.config.summaryBudget)
+      const history = buildConversationHistory(messages)
+
+      const t0 = Date.now()
+      summary = await callSummarizer(upstream, apiKey, prompt, history, this.config.summarizationTimeoutMs, undefined, profile)
+      logger.info('[export-compressor] full-llm done in %dms, %d chars', Date.now() - t0, summary!.length)
+    } catch (err: any) {
+      logger.warn('[export-compressor] full-llm failed: %s', err.message)
+    }
+
+    if (!summary) {
+      return { messages, meta }
+    }
+
+    return {
+      messages: [{ role: 'user', content: summary }],
+      meta: {
+        ...meta,
+        compressed: true,
+        llmCompressed: true,
+        summaryTokenEstimate: countTokens(summary),
+        verbatimCount: 0,
+      },
+    }
+  }
+}

packages/server/src/lib/context-compressor/index.ts

@@ -172,7 +172,7 @@ Be specific with file paths, commands, line numbers, and results.]
 ## Critical Context
 [Any specific values, error messages, configuration details, or data that would be lost without explicit preservation]`
 
-function buildFullPrompt(contentToSummarize: string, summaryBudget: number): string {
+export function buildFullPrompt(contentToSummarize: string, summaryBudget: number): string {
   return `You are a summarization agent creating a context checkpoint.
 Your output will be injected as reference material for a DIFFERENT
 assistant that continues the conversation.
@@ -194,7 +194,7 @@ Target ~${summaryBudget} tokens. Be CONCRETE — include file paths, command out
 Write only the summary body. Do not include any preamble or prefix.`
 }
 
-function buildIncrementalPrompt(previousSummary: string, contentToSummarize: string, summaryBudget: number): string {
+export function buildIncrementalPrompt(previousSummary: string, contentToSummarize: string, summaryBudget: number): string {
   return `You are a summarization agent creating a context checkpoint.
 Your output will be injected as reference material for a DIFFERENT
 assistant that continues the conversation.
@@ -229,7 +229,7 @@ Write only the summary body. Do not include any preamble or prefix.`
 
 // ─── Pre-cleaning ───────────────────────────────────────
 
-function serializeForSummary(messages: ChatMessage[]): string {
+export function serializeForSummary(messages: ChatMessage[]): string {
   const parts: string[] = []
 
   function contentToString(content: string | ContentBlock[]): string {
@@ -272,13 +272,13 @@ function serializeForSummary(messages: ChatMessage[]): string {
  * Convert messages to conversation history format for LLM API.
  * Tool calls are converted to text format within assistant messages.
  */
-function buildConversationHistory(messages: ChatMessage[]): Array<{ role: string; content: string }> {
+export function buildConversationHistory(messages: ChatMessage[]): Array<{ role: string; content: string }> {
   const result: Array<{ role: string; content: string }> = []
 
   for (const msg of messages) {
     if (msg.role === 'tool') {
       // Convert tool result to text and append to previous assistant message
-      const toolText = `[Tool result: ${msg.name || 'unknown'}]\n${(msg.content || '').slice(0, 500)}${msg.content && msg.content.length > 500 ? '...' : ''}`
+      const toolText = `[Tool result: ${msg.name || 'unknown'}]\n${(msg.content || '').slice(0, 4000)}${msg.content && msg.content.length > 4000 ? '...' : ''}`
       // Find the last assistant message and append to it
       const lastAssistant = result.findLast(m => m.role === 'assistant')
       if (lastAssistant) {
@@ -291,7 +291,7 @@ function buildConversationHistory(messages: ChatMessage[]): Array<{ role: string
       // Include tool calls in assistant message
       const toolsInfo = msg.tool_calls.map(tc => {
         let args = tc.function.arguments
-        if (args.length > 1000) args = args.slice(0, 1000) + '...'
+        if (args.length > 4000) args = args.slice(0, 4000) + '...'
         return `[Calling tool: ${tc.function.name} with arguments: ${args}]`
       }).join('\n')
       const content = msg.content ? `${msg.content}\n\n${toolsInfo}` : toolsInfo
@@ -313,6 +313,7 @@ function buildConversationHistory(messages: ChatMessage[]): Array<{ role: string
           }
         }
       }
+      if (contentStr.length > 4000) contentStr = contentStr.slice(0, 4000) + '...'
       result.push({ role: 'user', content: contentStr })
     } else if (msg.role === 'assistant' || msg.role === 'system') {
       let contentStr = ''
@@ -330,6 +331,7 @@ function buildConversationHistory(messages: ChatMessage[]): Array<{ role: string
           }
         }
       }
+      if (contentStr.length > 4000) contentStr = contentStr.slice(0, 4000) + '...'
       result.push({ role: msg.role, content: contentStr })
     }
     // Skip other roles
@@ -338,7 +340,7 @@ function buildConversationHistory(messages: ChatMessage[]): Array<{ role: string
   return result
 }
 
-function pruneOldToolResults(messages: ChatMessage[], keepRecentCount: number): ChatMessage[] {
+export function pruneOldToolResults(messages: ChatMessage[], keepRecentCount: number): ChatMessage[] {
   if (messages.length <= keepRecentCount) return messages
 
   const tail = messages.slice(-keepRecentCount)
@@ -365,7 +367,7 @@ function pruneOldToolResults(messages: ChatMessage[], keepRecentCount: number):
 
 // ─── LLM Summarization ──────────────────────────────────
 
-async function callSummarizer(
+export async function callSummarizer(
   upstream: string,
   apiKey: string | undefined,
   prompt: string,