/**
* GoogleContextBuilder - Builds conversation context for Google Gemini
*
* Google format uses:
* - 'user' and 'model' roles (not 'assistant')
* - 'parts' array with text, functionCall, or functionResponse objects
* - thoughtSignature for thinking models (Gemini 3.0+)
*
* Follows Single Responsibility Principle - only handles Google format.
*/
import { IContextBuilder, LLMMessage, LLMToolCall, ToolExecutionResult, GoogleMessage, GooglePart } from './IContextBuilder';
import { ConversationData, ChatMessage, ToolCall } from '../../../types/chat/ChatTypes';
import { ReasoningPreserver } from '../../llm/adapters/shared/ReasoningPreserver';
export class GoogleContextBuilder implements IContextBuilder {
readonly provider = 'google';
/**
* Validate if a message should be included in LLM context
*/
private isValidForContext(msg: ChatMessage, isLastMessage: boolean): boolean {
if (msg.state === 'invalid' || msg.state === 'streaming') return false;
if (msg.role === 'user' && (!msg.content || !msg.content.trim())) return false;
if (msg.role === 'assistant') {
const hasContent = msg.content && msg.content.trim();
const hasToolCalls = msg.toolCalls && msg.toolCalls.length > 0;
if (!hasContent && !hasToolCalls && !isLastMessage) return false;
if (hasToolCalls && msg.toolCalls) {
const allHaveResults = msg.toolCalls.every((tc: ToolCall) =>
tc.result !== undefined || tc.error !== undefined
);
if (!allHaveResults) return false;
}
}
return true;
}
/**
* Build context from stored conversation
*/
buildContext(conversation: ConversationData, systemPrompt?: string): LLMMessage[] {
const messages: GoogleMessage[] = [];
// Note: Google uses systemInstruction separately, not in messages
// But we include it here for compatibility with the interface
if (systemPrompt) {
messages.push({ role: 'system', parts: [{ text: systemPrompt }] });
}
// Filter valid messages
const validMessages = conversation.messages.filter((msg, index) => {
const isLastMessage = index === conversation.messages.length - 1;
return this.isValidForContext(msg, isLastMessage);
});
validMessages.forEach((msg) => {
if (msg.role === 'user') {
if (msg.content && msg.content.trim()) {
messages.push({
role: 'user',
parts: [{ text: msg.content }]
});
}
} else if (msg.role === 'assistant') {
if (msg.toolCalls && msg.toolCalls.length > 0) {
// Build model message with thought signatures preserved
const modelMessage = ReasoningPreserver.buildGoogleModelMessageWithThinking(
msg.toolCalls.map((tc: ToolCall) => ({
...tc,
function: { name: tc.name || '', arguments: JSON.stringify(tc.parameters || {}) }
}))
);
messages.push(modelMessage as GoogleMessage);
// Function response parts
const functionResponseParts: GooglePart[] = msg.toolCalls.map((tc: ToolCall) => ({
functionResponse: {
name: tc.name || '',
response: tc.success
? (tc.result || {})
: { error: tc.error || 'Tool execution failed' }
}
}));
messages.push({ role: 'function', parts: functionResponseParts });
// If there's final content after tool execution, add it
if (msg.content && msg.content.trim()) {
messages.push({
role: 'model',
parts: [{ text: msg.content }]
});
}
} else {
if (msg.content && msg.content.trim()) {
messages.push({
role: 'model',
parts: [{ text: msg.content }]
});
}
}
}
});
return messages;
}
/**
* Build tool continuation for pingpong pattern
*/
buildToolContinuation(
userPrompt: string,
toolCalls: LLMToolCall[],
toolResults: ToolExecutionResult[],
previousMessages?: LLMMessage[],
_systemPrompt?: string
): LLMMessage[] {
const messages: GoogleMessage[] = [];
// Add previous conversation history (convert to Google format if needed)
if (previousMessages && previousMessages.length > 0) {
for (const msg of previousMessages) {
// Skip messages that are already in Google format
const googleMsg = msg as GoogleMessage;
if (googleMsg.parts) {
messages.push(googleMsg);
continue;
}
// Convert from simple message format
const simpleMsg = msg as { role: string; content?: string };
if (simpleMsg.role === 'user' && simpleMsg.content) {
messages.push({
role: 'user',
parts: [{ text: typeof simpleMsg.content === 'string' ? simpleMsg.content : JSON.stringify(simpleMsg.content) }]
});
} else if (simpleMsg.role === 'assistant' && simpleMsg.content) {
messages.push({
role: 'model',
parts: [{ text: typeof simpleMsg.content === 'string' ? simpleMsg.content : JSON.stringify(simpleMsg.content) }]
});
}
}
}
// Add the original user message
if (userPrompt) {
messages.push({
role: 'user',
parts: [{ text: userPrompt }]
});
}
// Build model message with thought signatures preserved
const modelMessage = ReasoningPreserver.buildGoogleModelMessageWithThinking(toolCalls);
messages.push(modelMessage as GoogleMessage);
// Add function response parts
const functionResponseParts: GooglePart[] = toolResults.map(result => ({
functionResponse: {
name: result.name || result.function?.name || '',
response: result.success
? (result.result || {})
: { error: result.error || 'Tool execution failed' }
}
}));
messages.push({
role: 'user', // Google uses 'user' role for function responses
parts: functionResponseParts
});
return messages;
}
/**
* Append tool execution to existing history (no user message added)
*/
appendToolExecution(
toolCalls: LLMToolCall[],
toolResults: ToolExecutionResult[],
previousMessages: LLMMessage[]
): LLMMessage[] {
const messages: GoogleMessage[] = [...(previousMessages as GoogleMessage[])];
// Build model message with thought signatures preserved
const modelMessage = ReasoningPreserver.buildGoogleModelMessageWithThinking(toolCalls);
messages.push(modelMessage as GoogleMessage);
// Add function response parts
const functionResponseParts: GooglePart[] = toolResults.map(result => ({
functionResponse: {
name: result.name || result.function?.name || '',
response: result.success
? (result.result || {})
: { error: result.error || 'Tool execution failed' }
}
}));
messages.push({
role: 'user',
parts: functionResponseParts
});
return messages;
}
}
