feat: unify LLM streaming with thinking display and Web/cache API shims

- Extract runSimpleLlmStream to @openchatlab/node-runtime for shared
  Electron/Server streaming with thinking_start/delta/end event handling
- Simplify Electron IPC ai.ts by delegating to shared stream function
- Add thinking/thinkingDone fields to LlmStreamChunk and preload types
- Add collapsible thinking process display in SQL LAB with auto-scroll
- Implement Web chatStream via SSE with proper error propagation
- Add /_web/cache/save-to-downloads and show-in-folder server routes
- Add cacheApi shim in Web registry for server-side file operations
- Increase SQL LAB maxTokens to 4096 to accommodate reasoning models
- Detect finishReason:length and show token limit error message
- Add i18n keys for thinking/thinkingProcess/errorTokenLimit (4 locales)

packages/node-runtime/src/ai/index.ts

@@ -132,6 +132,10 @@ export type { RemoteModel, FetchRemoteModelsResult, RemoteApiOptions } from './r
 // Re-exports from @mariozechner/pi-agent-core
 export type { AgentTool, AgentToolResult } from '@mariozechner/pi-agent-core'
 
+// LLM simple streaming
+export { runSimpleLlmStream } from './llm-stream'
+export type { LlmStreamChunk, RunSimpleLlmStreamOptions } from './llm-stream'
+
 // Re-exports from @mariozechner/pi-ai
 export { Type, completeSimple, streamSimple } from '@mariozechner/pi-ai'
 export type {

packages/node-runtime/src/ai/llm-stream.ts

@@ -0,0 +1,119 @@
+/**
+ * Shared LLM simple streaming — used by both Electron and Server.
+ *
+ * Wraps pi-ai's streamSimple with message preprocessing,
+ * event mapping, and error handling.
+ */
+
+import { streamSimple } from '@mariozechner/pi-ai'
+import type { Model as PiModel, Api as PiApi, Message as PiMessage } from '@mariozechner/pi-ai'
+
+export interface LlmStreamChunk {
+  content: string
+  isFinished: boolean
+  finishReason?: 'stop' | 'length' | 'error'
+  error?: string
+  /** When present, this chunk carries thinking/reasoning content rather than final text. */
+  thinking?: string
+  /** Signals the end of a thinking block (thinking is empty string, thinkingDone is true). */
+  thinkingDone?: boolean
+}
+
+export interface RunSimpleLlmStreamOptions {
+  messages: Array<{ role: string; content: string }>
+  apiKey: string
+  piModel: PiModel<PiApi>
+  temperature?: number
+  maxTokens?: number
+  onChunk: (chunk: LlmStreamChunk) => void
+  abortSignal?: AbortSignal
+}
+
+function toPiMessages(messages: Array<{ role: string; content: string }>, timestamp: number): PiMessage[] {
+  return messages.map((m) => ({
+    role: m.role as 'user' | 'assistant',
+    content: m.content,
+    timestamp,
+  })) as unknown as PiMessage[]
+}
+
+/**
+ * Run a simple LLM streaming call.
+ * Separates system message, streams text deltas via onChunk callback.
+ */
+export async function runSimpleLlmStream(options: RunSimpleLlmStreamOptions): Promise<void> {
+  const { messages, apiKey, piModel, temperature, maxTokens, onChunk, abortSignal } = options
+
+  const systemMsg = messages.find((m) => m.role === 'system')
+  const nonSystemMsgs = messages.filter((m) => m.role !== 'system')
+  const now = Date.now()
+
+  const eventStream = streamSimple(
+    piModel,
+    {
+      systemPrompt: systemMsg?.content,
+      messages: toPiMessages(nonSystemMsgs, now),
+    },
+    { apiKey, temperature, maxTokens, signal: abortSignal }
+  )
+
+  let hasTerminalChunk = false
+
+  try {
+    for await (const event of eventStream) {
+      if (abortSignal?.aborted) {
+        if (!hasTerminalChunk) {
+          hasTerminalChunk = true
+          onChunk({ content: '', isFinished: true, finishReason: 'stop' })
+        }
+        return
+      }
+
+      if (event.type === 'thinking_start') {
+        onChunk({ content: '', isFinished: false, thinking: '' })
+        continue
+      }
+
+      if (event.type === 'thinking_delta') {
+        onChunk({ content: '', isFinished: false, thinking: event.delta })
+        continue
+      }
+
+      if (event.type === 'thinking_end') {
+        onChunk({ content: '', isFinished: false, thinking: '', thinkingDone: true })
+        continue
+      }
+
+      if (event.type === 'text_delta') {
+        onChunk({ content: event.delta, isFinished: false })
+        continue
+      }
+
+      if (event.type === 'done') {
+        hasTerminalChunk = true
+        onChunk({ content: '', isFinished: true, finishReason: event.reason === 'length' ? 'length' : 'stop' })
+        return
+      }
+
+      if (event.type === 'error') {
+        hasTerminalChunk = true
+        const errorMsg =
+          event.error?.content
+            ?.filter((c) => c.type === 'text')
+            .map((c) => ('text' in c ? c.text : ''))
+            .join('') || 'Unknown LLM error'
+        onChunk({ content: '', isFinished: true, finishReason: 'error', error: errorMsg })
+        return
+      }
+    }
+
+    if (!hasTerminalChunk) {
+      onChunk({ content: '', isFinished: true, finishReason: 'stop' })
+    }
+  } catch (error) {
+    if (!hasTerminalChunk) {
+      const msg = error instanceof Error ? error.message : String(error)
+      onChunk({ content: '', isFinished: true, finishReason: 'error', error: msg })
+    }
+  }
+}

packages/node-runtime/src/index.ts

@@ -240,7 +240,8 @@ export type { TempDbMeta, ExportedSession } from './merger/temp-db'
 
 // Re-exports: @mariozechner/pi-agent-core & @mariozechner/pi-ai
 export type { AgentTool, AgentToolResult } from './ai'
-export { Type, completeSimple, streamSimple } from './ai'
+export { Type, completeSimple, streamSimple, runSimpleLlmStream } from './ai'
+export type { LlmStreamChunk, RunSimpleLlmStreamOptions } from './ai'
 export type { PiModel, PiApi, PiMessage, PiUsage, PiTextContent, PiAssistantMessage } from './ai'
 
 // ==================== RAG ====================