feat: upgrade LLM adapters to support Gemini 3, Claude 4.5, and GPT-5… #31
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -21,7 +21,7 @@ import { ARTError, ErrorCode } from '@/errors'; // Import ARTError and ErrorCode | |
| export interface GeminiAdapterOptions { | ||
| /** Your Google AI API key (e.g., from Google AI Studio). Handle securely. */ | ||
| apiKey: string; | ||
| /** The default Gemini model ID to use (e.g., 'gemini-3-flash', 'gemini-pro'). Defaults to 'gemini-3-flash' if not provided. */ | ||
| model?: string; | ||
| /** Optional: Override the base URL for the Google Generative AI API. */ | ||
| apiBaseUrl?: string; // Note: Not directly used by SDK basic setup | ||
|
|
@@ -48,10 +48,10 @@ function isRetryableError(error: any): boolean { | |
| // Check for common retryable error messages | ||
| const message = String(error?.message || error?.error?.message || '').toLowerCase(); | ||
| return message.includes('overloaded') || | ||
| message.includes('rate limit') || | ||
| message.includes('temporarily unavailable') || | ||
| message.includes('503') || | ||
| message.includes('429'); | ||
| } | ||
|
|
||
| /** | ||
|
|
@@ -120,7 +120,7 @@ export class GeminiAdapter implements ProviderAdapter { | |
| throw new Error('GeminiAdapter requires an apiKey in options.'); | ||
| } | ||
| this.apiKey = options.apiKey; | ||
| this.defaultModel = options.model || 'gemini-3-flash'; // Default to latest stable flash model | ||
| // Initialize the SDK | ||
| // Use correct constructor based on documentation | ||
| this.genAI = new GoogleGenAI({ apiKey: this.apiKey }); | ||
|
|
@@ -138,10 +138,11 @@ export class GeminiAdapter implements ProviderAdapter { | |
| * Handles both streaming and non-streaming requests based on `options.stream`. | ||
| * | ||
| * Thinking tokens (Gemini): | ||
| * - On supported Gemini models (e.g., `gemini-3-*`), you can enable thought output via `config.thinkingConfig`. | ||
| * - This adapter reads provider-specific flags from the call options: | ||
| * - `options.gemini.thinking.includeThoughts: boolean` — when `true`, requests thought (reasoning) output. | ||
| * - `options.gemini.thinking.thinkingBudget?: number` — optional token budget for thinking. | ||
| * - `options.gemini.thinking.thinkingLevel?: 'low' | 'minimal' | 'medium' | 'high'` — optional thinking level (Gemini 3+). | ||
| * - When enabled and supported, the adapter will attempt to differentiate thought vs response parts and set | ||
| * `StreamEvent.tokenType` accordingly: | ||
| * - For planning calls (`callContext === 'AGENT_THOUGHT'`): `AGENT_THOUGHT_LLM_THINKING` or `AGENT_THOUGHT_LLM_RESPONSE`. | ||
|
|
@@ -187,8 +188,11 @@ export class GeminiAdapter implements ProviderAdapter { | |
|
|
||
| // --- Format Payload for SDK --- | ||
| let contents: Content[]; | ||
| let systemInstruction: string | undefined; | ||
| try { | ||
| const translation = this.translateToGemini(prompt); | ||
| contents = translation.contents; | ||
| systemInstruction = translation.systemInstruction; | ||
| } catch (error: any) { | ||
| Logger.error(`Error translating ArtStandardPrompt to Gemini format: ${error.message}`, { error, threadId, traceId }); | ||
| // Immediately yield error and end if translation fails | ||
|
|
@@ -213,19 +217,33 @@ export class GeminiAdapter implements ProviderAdapter { | |
| delete generationConfig[key as keyof GenerationConfig] | ||
| ); | ||
| // Build optional thinking configuration from CallOptions (feature flag) | ||
| // Expecting shape: options.gemini?.thinking?.{ includeThoughts?: boolean; thinkingBudget?: number; thinkingLevel?: string } | ||
| const geminiThinking = (options as any)?.gemini?.thinking; | ||
| const includeThoughts: boolean = !!geminiThinking?.includeThoughts; | ||
| const thinkingBudget: number | undefined = geminiThinking?.thinkingBudget; | ||
| const thinkingLevel: string | undefined = geminiThinking?.thinkingLevel; | ||
|
|
||
| // Merge into a requestConfig that is leniently typed to allow SDK preview fields | ||
| const requestConfig: any = includeThoughts | ||
| ? { | ||
| ...generationConfig, | ||
| thinkingConfig: { | ||
| includeThoughts: true, | ||
| ...(thinkingBudget !== undefined ? { thinkingBudget } : {}), | ||
| ...(thinkingLevel !== undefined ? { thinkingLevel } : {}), | ||
| }, | ||
| } | ||
| : { ...generationConfig }; | ||
|
|
||
| // Support systemInstruction in the newer SDK pattern | ||
| const callConfig: any = { | ||
| model: modelToUse, | ||
| contents, | ||
| config: requestConfig, | ||
| }; | ||
| if (systemInstruction) { | ||
| callConfig.systemInstruction = systemInstruction; | ||
| } | ||
| // --- End Format Payload --- | ||
|
|
||
| Logger.debug(`Calling Gemini SDK with model ${modelToUse}, stream: ${!!stream}`, { threadId, traceId }); | ||
|
|
@@ -247,11 +265,7 @@ export class GeminiAdapter implements ProviderAdapter { | |
| // Let TypeScript infer the type of streamResult | ||
| // Use the new SDK pattern: genAI.models.generateContentStream with retry logic | ||
| const streamResult = await withRetry( | ||
| () => genAIInstance.models.generateContentStream(callConfig), | ||
| { | ||
| onRetry: (error, attempt, delayMs) => { | ||
| Logger.info(`Retrying Gemini stream call (attempt ${attempt})`, { threadId, traceId, delayMs }); | ||
|
|
@@ -336,11 +350,7 @@ export class GeminiAdapter implements ProviderAdapter { | |
| } else { | ||
| // Use the new SDK pattern: genAIInstance.models.generateContent with retry logic | ||
| const result: GenerateContentResponse = await withRetry( | ||
| () => genAIInstance.models.generateContent(callConfig), | ||
| { | ||
| onRetry: (error, attempt, delayMs) => { | ||
| Logger.info(`Retrying Gemini call (attempt ${attempt})`, { threadId, traceId, delayMs }); | ||
|
|
@@ -414,62 +424,50 @@ export class GeminiAdapter implements ProviderAdapter { | |
| } | ||
|
|
||
| /** | ||
| * Translates the provider-agnostic `ArtStandardPrompt` into the Gemini API's `Content[]` format | ||
| * and extracts any system instructions. | ||
| * | ||
| * Key translations: | ||
| * - `system` role: Extracted to `systemInstruction`. | ||
| * - `user` role: Maps to Gemini's `user` role. | ||
| * - `assistant` role: Maps to Gemini's `model` role. Handles text content and `tool_calls` (mapped to `functionCall`). | ||
| * - `tool_result` role: Maps to Gemini's `user` role with a `functionResponse` part. | ||
| * - `tool_request` role: Skipped (implicitly handled by `assistant`'s `tool_calls`). | ||
| * | ||
| * @private | ||
| * @param {ArtStandardPrompt} artPrompt - The input `ArtStandardPrompt` array. | ||
| * @returns {{ contents: Content[], systemInstruction?: string }} The Gemini formatted payload. | ||
| * @throws {ARTError} If translation encounters an issue. | ||
| */ | ||
| private translateToGemini(artPrompt: ArtStandardPrompt): { contents: Content[], systemInstruction?: string } { | ||
| const contents: Content[] = []; | ||
| let systemInstruction: string | undefined; | ||
|
|
||
| for (const message of artPrompt) { | ||
| let role: 'user' | 'model'; | ||
| const parts: Part[] = []; | ||
|
|
||
| switch (message.role) { | ||
| case 'system': | ||
| if (typeof message.content === 'string') { | ||
| systemInstruction = (systemInstruction ? systemInstruction + "\n\n" : "") + message.content; | ||
| } else { | ||
| Logger.warn(`GeminiAdapter: Ignoring non-string system prompt content.`, { content: message.content }); | ||
| } | ||
| continue; | ||
|
|
||
| case 'user': { | ||
| role = 'user'; | ||
| let userContent = ''; | ||
| if (typeof message.content === 'string') { | ||
| userContent = message.content; | ||
| } else { | ||
| Logger.warn(`GeminiAdapter: Stringifying non-string user content.`, { content: message.content }); | ||
| userContent = JSON.stringify(message.content); | ||
| } | ||
| parts.push({ text: userContent }); | ||
| break; | ||
| } | ||
|
|
||
| case 'assistant': | ||
| role = 'model'; | ||
|
|
@@ -528,21 +526,15 @@ export class GeminiAdapter implements ProviderAdapter { | |
| continue; | ||
| } | ||
|
|
||
| contents.push({ role, parts }); | ||
| } | ||
|
|
||
| // Gemini specific validation: Ensure conversation doesn't start with 'model' | ||
| if (contents.length > 0 && contents[0].role === 'model') { | ||
| Logger.warn("Gemini conversation history starts with 'model' role. Prepending a dummy 'user' turn.", { firstRole: contents[0].role }); | ||
| contents.unshift({ role: 'user', parts: [{ text: "(Initial context)" }] }); // Prepend a generic user turn | ||
| } | ||
|
|
||
| return { contents, systemInstruction }; | ||
| } | ||
| } | ||
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There appears to be a stray
2.on this line which seems to be a typo. It should be removed to maintain the clean formatting of the changelog.