Files
Netcatty/components/ai/hooks/useAIChatStreaming.ts
T

1039 lines
40 KiB
TypeScript

/**
* useAIChatStreaming — Encapsulates all streaming logic for the AI chat panel.
*
* Handles:
* - Catty agent streaming via Vercel AI SDK `streamText`
* - External agent streaming through official SDK backends
* - Text-delta batching via requestAnimationFrame
* - Abort controller management
* - Stream state tracking (per-session)
* - Error reporting
*/
import React, { useCallback, useEffect, useRef, useState } from 'react';
import { generateText, pruneMessages, streamText, stepCountIs, type ModelMessage } from 'ai';
import type {
AIPermissionMode,
AIToolIntegrationMode,
AISession,
ChatMessage,
ChatMessageAttachment,
ExternalAgentConfig,
ProviderAdvancedParams,
ProviderConfig,
WebSearchConfig,
} from '../../../infrastructure/ai/types';
import { isWebSearchReady } from '../../../infrastructure/ai/types';
import { buildSystemPrompt } from '../../../infrastructure/ai/cattyAgent/systemPrompt';
import {
CONTEXT_COMPACTION_SYSTEM_PROMPT,
DEFAULT_CONTEXT_WINDOW_TOKENS,
DEFAULT_PROTECT_RECENT_MESSAGES,
formatMessagesForCompaction,
estimateUnknownTokens,
keepRecentContextMessages,
prepareContextCompaction,
resolveContextWindow,
} from '../../../infrastructure/ai/contextCompaction';
import { createModelFromConfig } from '../../../infrastructure/ai/sdk/providers';
import { createCattyTools } from '../../../infrastructure/ai/sdk/tools';
import type { ExecutorContext } from '../../../infrastructure/ai/cattyAgent/executor';
import { getExternalAgentSdkBackend } from '../../../infrastructure/ai/managedAgents';
import { runSdkAgentTurn } from '../../../infrastructure/ai/sdkAgentAdapter';
import { classifyError } from '../../../infrastructure/ai/errorClassifier';
import { isSdkStreamStateError } from '../../../infrastructure/ai/shared/streamStateErrors';
import {
buildPromptWithTerminalSelectionAttachments,
isTerminalSelectionAttachment,
} from '../../../application/state/terminalSelectionAttachment';
import {
extractProviderContinuationFromRawChunk,
getOpenAIChatAssistantFieldsForHistoryMessage,
isProviderContinuationForSource,
mergeProviderContinuation,
normalizeProviderContinuationOptions,
withProviderContinuationSource,
type OpenAIChatAssistantFields,
type ProviderContinuation,
} from '../../../infrastructure/ai/providerContinuation';
import {
getNetcattyBridge,
generateId,
isToolResultError,
resolveUserSkillsContext,
toAssistantModelContent,
type AssistantContentPart,
type CattyProviderContinuationContext,
type DefaultTargetSessionHint,
type ErrorChunk,
type RawChunk,
type ReasoningChunk,
type StreamChunk,
type TerminalSessionInfo,
type TextDeltaChunk,
type ToolCallChunk,
type ToolResultChunk,
} from './aiChatStreamingSupport';
export { getNetcattyBridge } from './aiChatStreamingSupport';
export type { DefaultTargetSessionHint } from './aiChatStreamingSupport';
const sharedStreamingSessionIds = new Set<string>();
const sharedAbortControllers = new Map<string, AbortController>();
const streamingSubscribers = new Set<() => void>();
const OPENAI_CHAT_ASSISTANT_FIELDS = Symbol('netcatty.openAIChatAssistantFields');
type ModelMessageWithOpenAIChatFields = ModelMessage & {
[OPENAI_CHAT_ASSISTANT_FIELDS]?: OpenAIChatAssistantFields;
};
function emitStreamingStoreChange(): void {
streamingSubscribers.forEach(listener => {
try {
listener();
} catch (err) {
console.error('[AIChatStreaming] Failed to notify streaming subscriber:', err);
}
});
}
function rememberOpenAIChatAssistantFields(
message: ModelMessage,
fields: OpenAIChatAssistantFields | undefined,
fieldsByMessage: Map<ModelMessage, OpenAIChatAssistantFields | undefined>,
): void {
fieldsByMessage.set(message, fields);
(message as ModelMessageWithOpenAIChatFields)[OPENAI_CHAT_ASSISTANT_FIELDS] = fields;
}
function getRememberedOpenAIChatAssistantFields(
message: ModelMessage,
fieldsByMessage: Map<ModelMessage, OpenAIChatAssistantFields | undefined>,
): OpenAIChatAssistantFields | undefined {
if (fieldsByMessage.has(message)) return fieldsByMessage.get(message);
return (message as ModelMessageWithOpenAIChatFields)[OPENAI_CHAT_ASSISTANT_FIELDS];
}
function modelMessageHasToolCall(message: ModelMessage): boolean {
if (message.role !== 'assistant' || !Array.isArray(message.content)) return false;
return message.content.some((part) => part && typeof part === 'object' && (part as { type?: string }).type === 'tool-call');
}
function collectOpenAIChatAssistantFieldsForMessages(
messages: ModelMessage[],
fieldsByMessage: Map<ModelMessage, OpenAIChatAssistantFields | undefined>,
): Array<OpenAIChatAssistantFields | undefined> {
const fields: Array<OpenAIChatAssistantFields | undefined> = [];
let previousMessageWasTool = false;
for (const message of messages) {
const needsContinuationFields = message.role === 'assistant'
&& (modelMessageHasToolCall(message) || previousMessageWasTool);
if (needsContinuationFields) {
fields.push(getRememberedOpenAIChatAssistantFields(message, fieldsByMessage));
}
previousMessageWasTool = message.role === 'tool';
}
return fields;
}
// -------------------------------------------------------------------
// Hook parameters
// -------------------------------------------------------------------
export interface UseAIChatStreamingParams {
maxIterations: number;
addMessageToSession: (sessionId: string, message: ChatMessage) => void;
updateLastMessage: (sessionId: string, updater: (msg: ChatMessage) => ChatMessage) => void;
updateMessageById: (sessionId: string, messageId: string, updater: (msg: ChatMessage) => ChatMessage) => void;
}
// -------------------------------------------------------------------
// Hook return type
// -------------------------------------------------------------------
export interface UseAIChatStreamingReturn {
/** Set of session IDs currently streaming. */
streamingSessionIds: Set<string>;
/** Set or unset streaming state for a session. */
setStreamingForScope: (key: string, val: boolean) => void;
/** Ref to per-session abort controllers. */
abortControllersRef: React.MutableRefObject<Map<string, AbortController>>;
/** Process a Catty agent stream. */
processCattyStream: (
streamSessionId: string,
model: ReturnType<typeof createModelFromConfig>,
systemPrompt: string,
tools: ReturnType<typeof createCattyTools>,
sdkMessages: Array<ModelMessage>,
signal: AbortSignal,
currentAssistantMsgId: string,
advancedParams?: ProviderAdvancedParams,
continuationContext?: CattyProviderContinuationContext,
) => Promise<void>;
/** Send a message to the Catty agent (built-in). */
sendToCattyAgent: (
sessionId: string,
sendScopeKey: string,
trimmed: string,
abortController: AbortController,
currentSession: AISession | undefined,
assistantMsgId: string,
context: SendToCattyContext,
attachments?: ChatMessageAttachment[],
) => Promise<void>;
/** Send a message to an external SDK agent. */
sendToExternalAgent: (
sessionId: string,
trimmed: string,
agentConfig: ExternalAgentConfig,
abortController: AbortController,
attachedImages: Array<{ base64Data: string; mediaType: string; filename?: string; filePath?: string }>,
context: SendToExternalContext,
) => Promise<void>;
/** Report a streaming error to the chat. */
reportStreamError: (sessionId: string, abortSignal: AbortSignal, err: unknown) => void;
}
/** Context values needed by sendToCattyAgent that change frequently (avoids stale closures). */
export interface SendToCattyContext {
activeProvider: ProviderConfig | undefined;
activeModelId: string;
scopeType: 'terminal' | 'workspace';
scopeTargetId?: string;
scopeLabel?: string;
globalPermissionMode: AIPermissionMode;
commandBlocklist?: string[];
terminalSessions: TerminalSessionInfo[];
webSearchConfig?: WebSearchConfig | null;
getExecutorContext?: () => ExecutorContext;
autoTitleSession: (sessionId: string, text: string) => void;
titleText?: string;
selectedUserSkillSlugs?: string[];
}
/** Context values needed by sendToExternalAgent that change frequently. */
export interface SendToExternalContext {
existingSessionId?: string;
updateExternalSessionId?: (sessionId: string, externalSessionId: string | undefined) => void;
historyMessages?: Array<{ role: 'user' | 'assistant'; content: string }>;
terminalSessions: TerminalSessionInfo[];
defaultTargetSession?: DefaultTargetSessionHint;
providers: ProviderConfig[];
selectedAgentModel?: string;
toolIntegrationMode: AIToolIntegrationMode;
selectedUserSkillSlugs?: string[];
}
// -------------------------------------------------------------------
// Hook implementation
// -------------------------------------------------------------------
export function useAIChatStreaming({
maxIterations,
addMessageToSession,
updateLastMessage,
updateMessageById,
}: UseAIChatStreamingParams): UseAIChatStreamingReturn {
// Per-session streaming state (keyed by sessionId)
const [streamingSessionIds, setStreamingSessions] = useState<Set<string>>(
() => new Set(sharedStreamingSessionIds),
);
useEffect(() => {
const syncFromStore = () => {
setStreamingSessions(new Set(sharedStreamingSessionIds));
};
streamingSubscribers.add(syncFromStore);
syncFromStore();
return () => {
streamingSubscribers.delete(syncFromStore);
};
}, []);
const setStreamingForScope = useCallback((key: string, val: boolean) => {
const hadKey = sharedStreamingSessionIds.has(key);
if (val) {
sharedStreamingSessionIds.add(key);
} else {
sharedStreamingSessionIds.delete(key);
}
if (hadKey !== val) {
emitStreamingStoreChange();
}
}, []);
// Per-scope abort controllers
const abortControllersRef = useRef<Map<string, AbortController>>(sharedAbortControllers);
// -------------------------------------------------------------------
// reportStreamError
// -------------------------------------------------------------------
const reportStreamError = useCallback((
sessionId: string,
abortSignal: AbortSignal,
err: unknown,
) => {
if (abortSignal.aborted) return;
// Log the full unsanitized error for debugging
console.error('[AIChatSidePanel] Stream error (full):', err);
// Pass the raw error to classifyError so it can inspect structured
// fields (statusCode, responseBody) from APICallError and friends;
// string-coercing here would strip the metadata we need to detect
// 413 / HTML-error-page / parse-failure scenarios.
const errorInfo = classifyError(err);
updateLastMessage(sessionId, msg => ({
...msg,
statusText: '',
executionStatus: msg.executionStatus === 'running' ? 'failed' : msg.executionStatus,
}));
addMessageToSession(sessionId, {
id: generateId(),
role: 'assistant',
content: '',
errorInfo,
timestamp: Date.now(),
});
}, [updateLastMessage, addMessageToSession]);
// -------------------------------------------------------------------
// processCattyStream
// -------------------------------------------------------------------
const processCattyStream = useCallback(async (
streamSessionId: string,
model: ReturnType<typeof createModelFromConfig>,
systemPrompt: string,
tools: ReturnType<typeof createCattyTools>,
sdkMessages: Array<ModelMessage>,
signal: AbortSignal,
currentAssistantMsgId: string,
advancedParams?: ProviderAdvancedParams,
continuationContext?: CattyProviderContinuationContext,
): Promise<void> => {
const result = streamText({
model,
messages: sdkMessages,
system: systemPrompt,
tools,
stopWhen: stepCountIs(maxIterations),
abortSignal: signal,
includeRawChunks: true,
...(advancedParams?.maxTokens != null && { maxOutputTokens: advancedParams.maxTokens }),
...(advancedParams?.temperature != null && { temperature: advancedParams.temperature }),
...(advancedParams?.topP != null && { topP: advancedParams.topP }),
...(advancedParams?.frequencyPenalty != null && { frequencyPenalty: advancedParams.frequencyPenalty }),
...(advancedParams?.presencePenalty != null && { presencePenalty: advancedParams.presencePenalty }),
});
// Track the current assistant message ID so updates target the correct message
let activeMsgId = currentAssistantMsgId;
let lastAddedRole: 'assistant' | 'tool' = 'assistant';
const reader = result.fullStream.getReader();
// -- Text-delta batching: accumulate deltas and flush periodically --
let pendingText = '';
let rafId: number | null = null;
const ensureAssistantMessage = (): string => {
if (lastAddedRole !== 'tool') return activeMsgId;
const newId = generateId();
addMessageToSession(streamSessionId, {
id: newId,
role: 'assistant',
content: '',
timestamp: Date.now(),
});
activeMsgId = newId;
lastAddedRole = 'assistant';
return activeMsgId;
};
const updateAssistantContinuation = (
messageId: string,
continuation: ProviderContinuation | undefined,
thinkingText = '',
) => {
if (!continuation && !thinkingText) return;
const sourcedContinuation = withProviderContinuationSource(continuation, continuationContext?.source);
updateMessageById(streamSessionId, messageId, msg => {
const providerContinuation = mergeProviderContinuation(msg.providerContinuation, sourcedContinuation);
return {
...msg,
...(providerContinuation ? { providerContinuation } : {}),
...(thinkingText ? { thinking: (msg.thinking || '') + thinkingText } : {}),
};
});
};
const getOpenAIReasoningText = (continuation: ProviderContinuation | undefined): string => {
const reasoningContent = continuation?.openAIChatAssistantFields?.reasoning_content;
return typeof reasoningContent === 'string' ? reasoningContent : '';
};
const flushText = () => {
if (pendingText) {
const text = pendingText;
pendingText = '';
if (lastAddedRole === 'tool') {
const newId = generateId();
addMessageToSession(streamSessionId, {
id: newId,
role: 'assistant',
content: text,
timestamp: Date.now(),
});
activeMsgId = newId;
lastAddedRole = 'assistant';
} else {
updateMessageById(streamSessionId, activeMsgId, msg => ({
...msg,
content: msg.content + text,
}));
}
}
rafId = null;
};
const cancelPendingFlush = () => {
if (rafId !== null) {
cancelAnimationFrame(rafId);
rafId = null;
}
};
try {
while (true) {
const { done, value } = await reader.read();
if (done) break;
// Use the StreamChunk union for type narrowing instead of unsafe casts
const chunk = value as StreamChunk;
switch (chunk.type) {
case 'text':
case 'text-delta': {
const typedChunk = chunk as TextDeltaChunk;
const text = typedChunk.text ?? typedChunk.textDelta;
const providerOptions = normalizeProviderContinuationOptions(typedChunk.providerMetadata);
if (providerOptions) {
const messageId = ensureAssistantMessage();
updateAssistantContinuation(messageId, { textProviderOptions: providerOptions });
}
if (text) {
pendingText += text;
if (rafId === null) {
rafId = requestAnimationFrame(flushText);
}
}
break;
}
case 'reasoning':
case 'reasoning-start':
case 'reasoning-delta': {
cancelPendingFlush();
flushText();
const typedChunk = chunk as ReasoningChunk;
const rText = typedChunk.text ?? typedChunk.textDelta ?? typedChunk.delta ?? '';
const providerOptions = normalizeProviderContinuationOptions(typedChunk.providerMetadata);
const continuation = rText || providerOptions
? {
reasoningParts: [{
text: rText,
...(providerOptions ? { providerOptions } : {}),
}],
} satisfies ProviderContinuation
: undefined;
if (continuation || rText) {
const messageId = ensureAssistantMessage();
updateAssistantContinuation(messageId, continuation, rText);
}
break;
}
case 'raw': {
const typedChunk = chunk as RawChunk;
const continuation = extractProviderContinuationFromRawChunk(typedChunk.rawValue);
if (continuation) {
cancelPendingFlush();
flushText();
const messageId = ensureAssistantMessage();
updateAssistantContinuation(messageId, continuation, getOpenAIReasoningText(continuation));
}
break;
}
case 'reasoning-end':
case 'text-start':
case 'text-end':
case 'start':
case 'finish':
case 'start-step':
case 'finish-step':
break;
case 'tool-call': {
cancelPendingFlush();
flushText();
const typedChunk = chunk as ToolCallChunk;
const messageId = ensureAssistantMessage();
const providerOptions = normalizeProviderContinuationOptions(typedChunk.providerMetadata);
updateMessageById(streamSessionId, messageId, msg => ({
...msg,
toolCalls: [...(msg.toolCalls || []), {
id: typedChunk.toolCallId,
name: typedChunk.toolName,
arguments: (typedChunk.input ?? typedChunk.args) as Record<string, unknown>,
}],
executionStatus: 'running',
statusText: undefined,
}));
if (providerOptions) {
updateAssistantContinuation(messageId, {
toolCallProviderOptionsById: {
[typedChunk.toolCallId]: providerOptions,
},
});
}
break;
}
case 'tool-result': {
cancelPendingFlush();
flushText();
const typedChunk = chunk as ToolResultChunk;
// Mark the assistant message's tool execution as completed
updateMessageById(streamSessionId, activeMsgId, msg =>
msg.role === 'assistant' && msg.executionStatus === 'running'
? { ...msg, executionStatus: 'completed', statusText: undefined } : msg,
);
const toolOutput = typedChunk.output ?? typedChunk.result;
const toolError = isToolResultError(toolOutput);
addMessageToSession(streamSessionId, {
id: generateId(),
role: 'tool',
content: '',
toolResults: [{
toolCallId: typedChunk.toolCallId,
content: typeof toolOutput === 'string'
? toolOutput
: JSON.stringify(toolOutput),
isError: toolError,
}],
timestamp: Date.now(),
executionStatus: 'completed',
});
lastAddedRole = 'tool';
break;
}
// tool-approval-request is no longer handled here — approval is now
// inside the tool's execute function via the approvalGate module.
// The SDK may still emit this chunk type but we simply ignore it.
case 'error': {
const typedChunk = chunk as ErrorChunk;
// Internal SDK reasoning/text state-machine errors (e.g. a
// third-party Anthropic-compat backend like DeepSeek's
// `-v4-flash` streaming thinking deltas without first emitting
// the `reasoning-start` content-block signal) leak through
// fullStream once per orphan delta. They're not user-facing
// errors — and worse, surfacing one assistant message per
// event breaks tool_use/tool_result contiguity on the next
// turn, which the Anthropic backend then rejects as
// `messages.N: tool_use ids were found without tool_result
// blocks immediately after`. Filter them out at the chunk
// boundary: drop the placeholder assistant message and keep
// accepting subsequent chunks, so the rest of the stream
// (real text, tool calls, the genuine `finish`) lands intact.
if (isSdkStreamStateError(typedChunk.error)) {
console.warn('[Catty] suppressed SDK stream state error:', typedChunk.error);
break;
}
cancelPendingFlush();
flushText();
updateMessageById(streamSessionId, activeMsgId, msg => ({
...msg,
statusText: '',
executionStatus: msg.executionStatus === 'running' ? 'failed' : msg.executionStatus,
}));
addMessageToSession(streamSessionId, {
id: generateId(),
role: 'assistant',
content: '',
// Pass the raw error so classifyError can detect 413 / HTML /
// schema-parse scenarios via structured fields (statusCode,
// responseBody) instead of lossy string conversion.
errorInfo: classifyError(typedChunk.error),
timestamp: Date.now(),
});
break;
}
default:
break;
}
}
} finally {
cancelPendingFlush();
flushText();
reader.releaseLock();
}
return;
}, [maxIterations, addMessageToSession, updateMessageById]);
// -------------------------------------------------------------------
// sendToExternalAgent
// -------------------------------------------------------------------
const sendToExternalAgent = useCallback(async (
sessionId: string,
trimmed: string,
agentConfig: ExternalAgentConfig,
abortController: AbortController,
attachedImages: Array<{ base64Data: string; mediaType: string; filename?: string; filePath?: string }>,
context: SendToExternalContext,
) => {
const bridge = getNetcattyBridge();
const userSkillsContext = await resolveUserSkillsContext(
bridge,
trimmed,
context.selectedUserSkillSlugs,
);
const sdkBackend = getExternalAgentSdkBackend(agentConfig);
if (sdkBackend && bridge) {
const requestId = `sdk_${Date.now()}_${Math.random().toString(36).slice(2, 6)}`;
// Push terminal session metadata to MCP bridge
if (bridge?.aiMcpUpdateSessions) {
await bridge.aiMcpUpdateSessions(context.terminalSessions, sessionId);
}
// Mutable flag: set after tool-result, cleared when new assistant msg is created
let needsNewAssistantMsg = false;
const maybeCreateAssistantMsg = () => {
if (needsNewAssistantMsg) {
needsNewAssistantMsg = false;
addMessageToSession(sessionId, {
id: generateId(), role: 'assistant', content: '', timestamp: Date.now(),
model: agentConfig.name || 'external',
});
}
};
await runSdkAgentTurn(
bridge,
requestId,
sessionId,
agentConfig,
trimmed,
{
onTextDelta: (text: string) => {
maybeCreateAssistantMsg();
updateLastMessage(sessionId, msg => ({
...msg,
content: msg.content + text,
statusText: undefined,
thinkingDurationMs: msg.thinking && !msg.thinkingDurationMs
? Date.now() - msg.timestamp : msg.thinkingDurationMs,
}));
},
onThinkingDelta: (text: string) => {
maybeCreateAssistantMsg();
updateLastMessage(sessionId, msg => ({
...msg, thinking: (msg.thinking || '') + text,
}));
},
onThinkingDone: () => {
updateLastMessage(sessionId, msg => ({
...msg, thinkingDurationMs: msg.thinkingDurationMs || (Date.now() - msg.timestamp),
}));
},
onToolCall: (toolName: string, args: Record<string, unknown>, toolCallId?: string) => {
maybeCreateAssistantMsg();
updateLastMessage(sessionId, msg => ({
...msg,
toolCalls: [...(msg.toolCalls || []), { id: toolCallId || `tc_${Date.now()}`, name: toolName, arguments: args }],
executionStatus: 'running',
statusText: undefined,
}));
},
onToolResult: (toolCallId: string, result: string, toolName?: string) => {
updateLastMessage(sessionId, msg => {
if (msg.role !== 'assistant' || msg.executionStatus !== 'running') return msg;
// Only patch tool call name if the existing name is missing/generic
// (don't overwrite a good name from onToolCall with a wrapper name from tool-result)
const updatedToolCalls = toolName && !toolName.includes('sdk_agent_dynamic_tool') && msg.toolCalls
? msg.toolCalls.map(tc => tc.id === toolCallId && !tc.name ? { ...tc, name: toolName } : tc)
: msg.toolCalls;
return { ...msg, toolCalls: updatedToolCalls, executionStatus: 'completed', statusText: undefined };
});
const toolError = isToolResultError(result);
addMessageToSession(sessionId, {
id: generateId(), role: 'tool', content: '',
toolResults: [{ toolCallId, content: result, isError: toolError }],
timestamp: Date.now(), executionStatus: 'completed',
});
needsNewAssistantMsg = true;
},
onStatus: (message: string) => {
maybeCreateAssistantMsg();
updateLastMessage(sessionId, msg => ({ ...msg, statusText: message }));
},
onSessionId: (externalSessionId: string) => {
context.updateExternalSessionId?.(sessionId, externalSessionId);
},
onError: (error: string) => {
reportStreamError(sessionId, abortController.signal, error);
setStreamingForScope(sessionId, false);
},
onDone: () => {},
},
abortController.signal,
// Managed SDK agents (codex, claude) must resolve auth from their own
// CLI config/login state, so we deliberately pass no providerId here.
// See issue #705 for Codex; same reasoning for Claude.
undefined,
context.selectedAgentModel,
context.existingSessionId,
context.historyMessages,
attachedImages.length > 0 ? attachedImages : undefined,
context.toolIntegrationMode,
context.defaultTargetSession,
userSkillsContext,
);
} else {
// Managed agents always route through the SDK path above.
reportStreamError(
sessionId,
abortController.signal,
'This agent has no SDK backend configured. Re-discover it in Settings -> AI.',
);
setStreamingForScope(sessionId, false);
}
}, [
addMessageToSession, updateLastMessage, setStreamingForScope, reportStreamError,
]);
// -------------------------------------------------------------------
// sendToCattyAgent
// -------------------------------------------------------------------
const sendToCattyAgent = useCallback(async (
sessionId: string,
sendScopeKey: string,
trimmed: string,
abortController: AbortController,
currentSession: AISession | undefined,
assistantMsgId: string,
context: SendToCattyContext,
attachments?: ChatMessageAttachment[],
) => {
const bridge = getNetcattyBridge();
const userSkillsContext = await resolveUserSkillsContext(
bridge,
trimmed,
context.selectedUserSkillSlugs,
);
const getExecutorContext = context.getExecutorContext ?? (() => ({
sessions: context.terminalSessions,
workspaceId: context.scopeType === 'workspace' ? context.scopeTargetId : undefined,
workspaceName: context.scopeType === 'workspace' ? context.scopeLabel : undefined,
}));
const tools = createCattyTools(
bridge,
getExecutorContext,
context.commandBlocklist,
context.globalPermissionMode,
context.webSearchConfig ?? undefined,
sessionId,
);
const systemPrompt = buildSystemPrompt({
scopeType: context.scopeType, scopeLabel: context.scopeLabel,
hosts: context.terminalSessions.map(s => ({
sessionId: s.sessionId, hostname: s.hostname, label: s.label,
os: s.os,
username: s.username,
protocol: s.protocol,
shellType: s.shellType,
deviceType: s.deviceType,
connected: s.connected,
})),
permissionMode: context.globalPermissionMode,
webSearchEnabled: isWebSearchReady(context.webSearchConfig),
userSkillsContext,
});
// Guard: activeProvider must exist for Catty agent path
if (!context.activeProvider) {
reportStreamError(sessionId, abortController.signal, 'No AI provider configured. Please configure a provider in Settings → AI.');
return;
}
const activeModelId = context.activeModelId || context.activeProvider.defaultModel || '';
const continuationContext: CattyProviderContinuationContext = {
source: {
providerConfigId: context.activeProvider.id,
providerType: context.activeProvider.providerId,
modelId: activeModelId,
},
openAIChatAssistantFields: [],
};
try {
// Issue #5: Build SDK messages including tool-call and tool-result messages
// so the LLM maintains full conversation context
const allMessages = currentSession?.messages ?? [];
// Collect all tool call IDs that have a corresponding tool result,
// so we can skip orphaned tool calls (e.g. from user stopping mid-execution)
const resolvedToolCallIds = new Set<string>();
for (const m of allMessages) {
if (m.role === 'tool' && m.toolResults) {
for (const tr of m.toolResults) resolvedToolCallIds.add(tr.toolCallId);
}
}
const findToolName = (toolCallId: string): string => {
for (const prev of allMessages) {
if (prev.role === 'assistant' && prev.toolCalls) {
const tc = prev.toolCalls.find(t => t.id === toolCallId);
if (tc) return tc.name;
}
}
return 'unknown';
};
const sdkMessages: Array<ModelMessage> = [];
const openAIChatAssistantFieldsByMessage = new Map<ModelMessage, OpenAIChatAssistantFields | undefined>();
let previousHistoryMessageWasToolResult = false;
for (const m of allMessages) {
const currentMessageFollowsToolResult = previousHistoryMessageWasToolResult;
if (m.role === 'user') {
// Build multimodal content when attachments are present (fallback to legacy `images` field)
const messageAttachments = m.attachments ?? m.images;
const modelText = messageAttachments?.length
? buildPromptWithTerminalSelectionAttachments(m.content, messageAttachments)
: m.content;
const modelAttachments = messageAttachments?.filter(
(attachment) => !isTerminalSelectionAttachment(attachment),
);
if (modelAttachments?.length) {
const parts: Array<{ type: 'text'; text: string } | { type: 'image'; image: string; mediaType?: string } | { type: 'file'; data: string; mediaType: string; filename?: string }> = [];
parts.push({ type: 'text', text: modelText });
for (const att of modelAttachments) {
if (att.mediaType.startsWith('image/')) {
parts.push({ type: 'image', image: att.base64Data, mediaType: att.mediaType });
} else {
parts.push({ type: 'file', data: att.base64Data, mediaType: att.mediaType, filename: att.filename });
}
}
sdkMessages.push({ role: 'user', content: parts });
} else {
sdkMessages.push({ role: 'user', content: modelText });
}
} else if (m.role === 'assistant') {
const activeContinuation = isProviderContinuationForSource(
m.providerContinuation,
continuationContext.source,
)
? m.providerContinuation
: undefined;
const openAIChatAssistantFields = getOpenAIChatAssistantFieldsForHistoryMessage(
m,
continuationContext.source,
);
if (m.toolCalls?.length) {
// Only include tool calls that have matching results
const resolvedCalls = m.toolCalls.filter(tc => resolvedToolCallIds.has(tc.id));
const contentParts: AssistantContentPart[] = [];
if (resolvedCalls.length > 0) {
for (const part of activeContinuation?.reasoningParts ?? []) {
if (!part.text && !part.providerOptions) continue;
contentParts.push({
type: 'reasoning' as const,
text: part.text,
...(part.providerOptions ? { providerOptions: part.providerOptions } : {}),
});
}
}
if (m.content) {
contentParts.push({
type: 'text' as const,
text: m.content,
...(activeContinuation?.textProviderOptions ? { providerOptions: activeContinuation.textProviderOptions } : {}),
});
}
for (const tc of resolvedCalls) {
const providerOptions = activeContinuation?.toolCallProviderOptionsById?.[tc.id];
contentParts.push({
type: 'tool-call' as const,
toolCallId: tc.id,
toolName: tc.name,
input: tc.arguments ?? {},
...(providerOptions ? { providerOptions } : {}),
});
}
// If all tool calls were orphaned, just include the text content
if (contentParts.length > 0) {
const message: ModelMessage = { role: 'assistant', content: toAssistantModelContent(contentParts) };
sdkMessages.push(message);
if (resolvedCalls.length > 0) {
rememberOpenAIChatAssistantFields(message, openAIChatAssistantFields, openAIChatAssistantFieldsByMessage);
}
}
} else if (m.content) {
const contentParts: AssistantContentPart[] = [];
for (const part of activeContinuation?.reasoningParts ?? []) {
if (!part.text && !part.providerOptions) continue;
contentParts.push({
type: 'reasoning' as const,
text: part.text,
...(part.providerOptions ? { providerOptions: part.providerOptions } : {}),
});
}
contentParts.push({
type: 'text' as const,
text: m.content,
...(activeContinuation?.textProviderOptions ? { providerOptions: activeContinuation.textProviderOptions } : {}),
});
const message: ModelMessage = {
role: 'assistant',
content: toAssistantModelContent(contentParts),
};
sdkMessages.push(message);
if (currentMessageFollowsToolResult) {
rememberOpenAIChatAssistantFields(message, openAIChatAssistantFields, openAIChatAssistantFieldsByMessage);
}
}
} else if (m.role === 'tool' && m.toolResults?.length) {
sdkMessages.push({
role: 'tool',
content: m.toolResults.map(tr => ({
type: 'tool-result' as const,
toolCallId: tr.toolCallId,
toolName: findToolName(tr.toolCallId),
output: { type: 'text' as const, value: tr.content },
})),
});
}
previousHistoryMessageWasToolResult = m.role === 'tool' && !!m.toolResults?.length;
}
// Build the current user message — include attachments as multimodal content
if (attachments?.length) {
const parts: Array<{ type: 'text'; text: string } | { type: 'image'; image: string; mediaType?: string } | { type: 'file'; data: string; mediaType: string; filename?: string }> = [];
parts.push({ type: 'text', text: trimmed });
for (const att of attachments) {
if (att.mediaType.startsWith('image/')) {
parts.push({ type: 'image', image: att.base64Data, mediaType: att.mediaType });
} else {
parts.push({ type: 'file', data: att.base64Data, mediaType: att.mediaType, filename: att.filename });
}
}
sdkMessages.push({ role: 'user', content: parts });
} else {
sdkMessages.push({ role: 'user', content: trimmed });
}
// Create model with placeholder API key — the main process injects the real
// decrypted key when the HTTP request is proxied through IPC, so plaintext
// keys never transit the renderer ↔ main IPC boundary.
let model;
try {
model = createModelFromConfig(
{
...context.activeProvider,
defaultModel: activeModelId,
},
{
getOpenAIChatAssistantFields: () => continuationContext.openAIChatAssistantFields,
},
);
} catch (e) {
console.error('[Catty] Model creation failed:', e);
reportStreamError(sessionId, abortController.signal, `Model creation failed: ${e instanceof Error ? e.message : String(e)}`);
return;
}
const contextWindow = resolveContextWindow({
provider: context.activeProvider,
modelId: activeModelId,
defaultContextWindow: DEFAULT_CONTEXT_WINDOW_TOKENS,
});
const outputReserveTokens = Math.min(4096, Math.ceil(contextWindow * 0.05));
const requestReserveTokens = outputReserveTokens + estimateUnknownTokens({
systemPrompt,
toolNames: Object.keys(tools),
openAIChatAssistantFields: Array.from(openAIChatAssistantFieldsByMessage.values()),
});
let messagesForStream = sdkMessages;
try {
const compacted = await prepareContextCompaction({
messages: sdkMessages,
contextWindow,
reservedTokens: requestReserveTokens,
protectRecentMessages: DEFAULT_PROTECT_RECENT_MESSAGES,
summarize: async (messagesToSummarize) => {
updateLastMessage(sessionId, msg => ({ ...msg, statusText: 'Compacting earlier context...' }));
const result = await generateText({
model,
system: CONTEXT_COMPACTION_SYSTEM_PROMPT,
messages: [{
role: 'user',
content: `Summarize this earlier conversation context for the next model turn:\n\n${formatMessagesForCompaction(messagesToSummarize)}`,
}],
abortSignal: abortController.signal,
maxOutputTokens: 1600,
temperature: 0,
});
return result.text;
},
});
messagesForStream = compacted.messages;
} catch (err) {
if (abortController.signal.aborted) throw err;
console.warn('[Catty] Context compaction failed; falling back to recent messages only:', err);
messagesForStream = keepRecentContextMessages(sdkMessages, DEFAULT_PROTECT_RECENT_MESSAGES);
}
messagesForStream = pruneMessages({
messages: messagesForStream,
reasoning: 'all',
emptyMessages: 'remove',
});
continuationContext.openAIChatAssistantFields = collectOpenAIChatAssistantFieldsForMessages(
messagesForStream,
openAIChatAssistantFieldsByMessage,
);
await processCattyStream(
sessionId,
model,
systemPrompt,
tools,
messagesForStream,
abortController.signal,
assistantMsgId,
context.activeProvider?.advancedParams,
continuationContext,
);
} catch (err) {
console.error('[Catty] streamText error:', err);
reportStreamError(sessionId, abortController.signal, err);
} finally {
// Clear any lingering statusText when the stream finishes
updateLastMessage(sessionId, msg => msg.statusText ? { ...msg, statusText: '' } : msg);
setStreamingForScope(sessionId, false);
abortControllersRef.current.delete(sessionId);
context.autoTitleSession(sessionId, context.titleText ?? trimmed);
}
}, [
processCattyStream, reportStreamError, setStreamingForScope,
updateLastMessage,
]);
return {
streamingSessionIds,
setStreamingForScope,
abortControllersRef,
processCattyStream,
sendToCattyAgent,
sendToExternalAgent,
reportStreamError,
};
}