shumengya/sproutclaw
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(ai): expose provider response ids on assistant messages fixes #2245

Browse Source
This commit is contained in:
Mario Zechner
2026-03-17 17:11:18 +01:00
parent 18d90b5c48
commit dd53eb56ee
11 changed files with 174 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
packages/ai/src/providers/anthropic.ts
View File

@@ -264,6 +264,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
for await (const event of anthropicStream) {
if (event.type === "message_start") {
output.responseId = event.message.id;
// Capture initial token usage from message_start event
// This ensures we have input token counts even if the stream is aborted early
output.usage.input = event.message.usage.input_tokens || 0;

3
packages/ai/src/providers/google-gemini-cli.ts
View File

@@ -548,6 +548,9 @@ export const streamGoogleGeminiCli: StreamFunction<"google-gemini-cli", GoogleGe
// Unwrap the response
const responseData = chunk.response;
if (!responseData) continue;
// Cloud Code Assist mirrors Gemini's responseId field. Keep the first non-empty one.
// A single streamed response should retain the same ID across chunks.
output.responseId ||= responseData.responseId;
const candidate = responseData.candidates?.[0];
if (candidate?.content?.parts) {

3
packages/ai/src/providers/google-vertex.ts
View File

@@ -101,6 +101,9 @@ export const streamGoogleVertex: StreamFunction<"google-vertex", GoogleVertexOpt
const blocks = output.content;
const blockIndex = () => blocks.length - 1;
for await (const chunk of googleStream) {
// Vertex uses the same @google/genai GenerateContentResponse type as Gemini.
// responseId is documented there as an output-only identifier for each response.
output.responseId ||= chunk.responseId;
const candidate = chunk.candidates?.[0];
if (candidate?.content?.parts) {
for (const part of candidate.content.parts) {

3
packages/ai/src/providers/google.ts
View File

@@ -86,6 +86,9 @@ export const streamGoogle: StreamFunction<"google-generative-ai", GoogleOptions>
const blocks = output.content;
const blockIndex = () => blocks.length - 1;
for await (const chunk of googleStream) {
// @google/genai documents GenerateContentResponse.responseId as an output-only field
// used to identify each response. Keep the first non-empty one from the stream.
output.responseId ||= chunk.responseId;
const candidate = chunk.candidates?.[0];
if (candidate?.content?.parts) {
for (const part of candidate.content.parts) {

3
packages/ai/src/providers/mistral.ts
View File

@@ -285,6 +285,9 @@ async function consumeChatStream(
for await (const event of mistralStream) {
const chunk = event.data;
// Mistral's streamed CompletionChunk carries an id field. Keep the first non-empty one,
// mirroring how OpenAI-style streaming exposes a stable response identifier per stream.
output.responseId ||= chunk.id;
if (chunk.usage) {
output.usage.input = chunk.usage.promptTokens || 0;

3
packages/ai/src/providers/openai-completions.ts
View File

@@ -127,6 +127,9 @@ export const streamOpenAICompletions: StreamFunction<"openai-completions", OpenA
};
for await (const chunk of openaiStream) {
// OpenAI documents ChatCompletionChunk.id as the unique chat completion identifier,
// and each chunk in a streamed completion carries the same id.
output.responseId ||= chunk.id;
if (chunk.usage) {
output.usage = parseChunkUsage(chunk.usage, model);
}

7
packages/ai/src/providers/openai-responses-shared.ts
View File

@@ -285,7 +285,9 @@ export async function processResponsesStream<TApi extends Api>(
const blockIndex = () => blocks.length - 1;
for await (const event of openaiStream) {
if (event.type === "response.output_item.added") {
if (event.type === "response.created") {
output.responseId = event.response.id;
} else if (event.type === "response.output_item.added") {
const item = event.item;
if (item.type === "reasoning") {
currentItem = item;
@@ -442,6 +444,9 @@ export async function processResponsesStream<TApi extends Api>(
}
} else if (event.type === "response.completed") {
const response = event.response;
if (response?.id) {
output.responseId = response.id;
}
if (response?.usage) {
const cachedTokens = response.usage.input_tokens_details?.cached_tokens || 0;
output.usage = {

1
packages/ai/src/types.ts
View File

@@ -193,6 +193,7 @@ export interface AssistantMessage {
api: Api;
provider: Provider;
model: string;
responseId?: string; // Provider-specific response/message identifier when the upstream API exposes one
usage: Usage;
stopReason: StopReason;
errorMessage?: string;