Gemini vercel ai sdk adapter #56
Merged
Conversation
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
Contributor
Greptile OverviewGreptile SummaryThis PR adds a comprehensive Vercel AI SDK adapter that enables BrowserOS to use multiple LLM providers (OpenAI, Anthropic, Google, Azure, Bedrock, OpenRouter, Ollama, LMStudio) through a unified interface. The implementation uses a clean strategy pattern for converting between Gemini and Vercel AI SDK formats. Key additions:
Issues found:
Confidence Score: 4/5
Important Files ChangedFile Analysis
Sequence DiagramsequenceDiagram
participant Client as Client/CLI
participant Adapter as VercelAIContentGenerator
participant MsgStrategy as MessageConversionStrategy
participant ToolStrategy as ToolConversionStrategy
participant RespStrategy as ResponseConversionStrategy
participant Provider as AI Provider (OpenAI/Anthropic/etc)
participant Stream as HonoSSEStream
Client->>Adapter: generateContentStream(request)
Adapter->>MsgStrategy: geminiToVercel(contents)
MsgStrategy-->>Adapter: CoreMessage[]
Adapter->>ToolStrategy: geminiToVercel(tools)
ToolStrategy->>ToolStrategy: normalizeForOpenAI(schema)
ToolStrategy-->>Adapter: VercelTool[]
Adapter->>Provider: streamText({messages, tools, system})
Provider-->>Adapter: fullStream (AsyncIterable)
loop For each stream chunk
Provider->>RespStrategy: chunk (text-delta/tool-call/finish)
alt text-delta chunk
RespStrategy->>Stream: write SSE (text-delta)
RespStrategy-->>Adapter: yield GenerateContentResponse
else tool-call chunk
RespStrategy->>Stream: write SSE (tool-call)
RespStrategy->>RespStrategy: accumulate in toolCallsMap
else finish chunk
RespStrategy->>RespStrategy: store finishReason
end
end
RespStrategy->>Provider: getUsage()
Provider-->>RespStrategy: usage metadata
RespStrategy->>Stream: write SSE (finish event)
RespStrategy->>ToolStrategy: vercelToGemini(toolCalls)
ToolStrategy-->>RespStrategy: FunctionCall[]
RespStrategy-->>Adapter: yield final GenerateContentResponse
Adapter-->>Client: AsyncGenerator<GenerateContentResponse>
|
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
Comment on lines
40
to
46
| const propertyKeys = Object.keys(result.properties); | ||
| if (propertyKeys.length > 0) { | ||
| // Merge with existing required array (if any) and ensure all keys are included | ||
| const existingRequired = Array.isArray(result.required) ? result.required : []; | ||
| const allRequired = Array.from(new Set([...existingRequired, ...propertyKeys])); | ||
| result.required = allRequired; | ||
| } |
Contributor
There was a problem hiding this comment.
logic: making all properties required may break tools with optional parameters
the normalization adds all property keys to required array, making every parameter mandatory. tools with genuinely optional params will fail if those params aren't provided.
Suggested change
| const propertyKeys = Object.keys(result.properties); | |
| if (propertyKeys.length > 0) { | |
| // Merge with existing required array (if any) and ensure all keys are included | |
| const existingRequired = Array.isArray(result.required) ? result.required : []; | |
| const allRequired = Array.from(new Set([...existingRequired, ...propertyKeys])); | |
| result.required = allRequired; | |
| } | |
| // 2. Add required array with ALL property keys | |
| if (result.properties && typeof result.properties === 'object') { | |
| const propertyKeys = Object.keys(result.properties); | |
| if (propertyKeys.length > 0) { | |
| // Only merge with existing required array, don't force all properties to be required | |
| if (Array.isArray(result.required) && result.required.length > 0) { | |
| result.required = Array.from(new Set(result.required)); | |
| } | |
| } | |
| } |
Prompt To Fix With AI
This is a comment left during a code review.
Path: packages/agent/src/agent/gemini-vercel-sdk-adapter/strategies/tool.ts
Line: 40:46
Comment:
**logic:** making all properties required may break tools with optional parameters
the normalization adds all property keys to `required` array, making every parameter mandatory. tools with genuinely optional params will fail if those params aren't provided.
```suggestion
// 2. Add required array with ALL property keys
if (result.properties && typeof result.properties === 'object') {
const propertyKeys = Object.keys(result.properties);
if (propertyKeys.length > 0) {
// Only merge with existing required array, don't force all properties to be required
if (Array.isArray(result.required) && result.required.length > 0) {
result.required = Array.from(new Set(result.required));
}
}
}
```
How can I resolve this? If you propose a fix, please make it concise.
felarof99
reviewed
Nov 25, 2025
| /** | ||
| * Configuration for Vercel AI adapter | ||
| */ | ||
| export interface VercelAIConfig { |
Contributor
There was a problem hiding this comment.
@greptileai what is a better way to have this structure? Can we have anthropic, openai, google as enums? and have a different structure for the config?
Contributor
There was a problem hiding this comment.
Looking at the current VercelAIConfig interface, I can suggest a better structure that uses enums for providers and organizes the configuration more logically:
Suggested change
| export interface VercelAIConfig { | |
| /** | |
| * Supported AI providers | |
| */ | |
| export enum AIProvider { | |
| ANTHROPIC = 'anthropic', | |
| OPENAI = 'openai', | |
| GOOGLE = 'google', | |
| OPENROUTER = 'openrouter', | |
| AZURE = 'azure', | |
| OLLAMA = 'ollama', | |
| LMSTUDIO = 'lmstudio', | |
| AWS_BEDROCK = 'aws-bedrock', | |
| } | |
| /** | |
| * Provider-specific configuration | |
| */ | |
| export interface ProviderConfig { | |
| apiKey?: string; | |
| baseUrl?: string; | |
| region?: string; | |
| resourceName?: string; | |
| accessKeyId?: string; | |
| secretAccessKey?: string; | |
| sessionToken?: string; | |
| } | |
| /** | |
| * Configuration for Vercel AI adapter | |
| */ | |
| export interface VercelAIConfig { | |
| model: string; | |
| provider: AIProvider; | |
| providerConfig?: ProviderConfig; | |
| honoStream?: HonoSSEStream; | |
| } |
This structure provides several benefits:
- Type safety: Using enums prevents typos and provides autocomplete
- Cleaner organization: All provider-specific settings are grouped under
providerConfig - Extensibility: Easy to add new providers without cluttering the main interface
- Explicit provider selection: Forces users to specify which provider they're using
- Flexible configuration: The
ProviderConfigcan accommodate different providers' needs without optional fields for every possible provider
This way, instead of having separate optional fields for each provider's configuration scattered in the main interface, everything is properly organized and type-safe.
Contributor
There was a problem hiding this comment.
@shivammittal274 could you update the code to use something similar to above design?
Contributor
|
LGTM! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
No description provided.