@oh-my-pi/pi-ai 16.5.2 → 17.0.0

package/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [17.0.0] - 2026-07-15
+
+### Changed
+
+- Improved Ollama streaming performance by parsing NDJSON response bytes directly instead of decoding and buffering network chunks as text.
+
+### Fixed
+
+- Fixed Cursor TLS connection resets causing process-fatal uncaught exceptions, allowing the active turn to fail or retry gracefully without terminating the session.
+- Fixed Amazon Bedrock stream error handling to correctly handle non-Error values that cannot be serialized by JSON.stringify.
+
 ## [16.5.2] - 2026-07-14
 
 ### Added

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "16.5.2",
+	"version": "17.0.0",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -38,9 +38,9 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.1",
-		"@oh-my-pi/pi-catalog": "16.5.2",
-		"@oh-my-pi/pi-utils": "16.5.2",
-		"@oh-my-pi/pi-wire": "16.5.2",
+		"@oh-my-pi/pi-catalog": "17.0.0",
+		"@oh-my-pi/pi-utils": "17.0.0",
+		"@oh-my-pi/pi-wire": "17.0.0",
 		"arktype": "2.2.3",
 		"zod": "^4"
 	},

package/src/providers/amazon-bedrock.ts

@@ -510,7 +510,12 @@ export const streamBedrock: StreamFunction<"bedrock-converse-stream"> = (
 			for (const block of output.content) {
 				if (block.type === "toolCall") clearStreamingPartialJson(block);
 			}
-			const baseMessage = error instanceof Error ? error.message : JSON.stringify(error);
+			let baseMessage: string;
+			try {
+				baseMessage = error instanceof Error ? error.message : (JSON.stringify(error) ?? String(error));
+			} catch {
+				baseMessage = String(error);
+			}
 			// Enrich error with thinking block diagnostics for signature-related failures
 			let diagnostics = "";
 			if (baseMessage.includes("signature") || baseMessage.includes("thinking")) {

package/src/providers/cursor.ts

@@ -351,6 +351,8 @@ export const streamCursor: StreamFunction<"cursor-agent"> = (
 		let h2Request: http2.ClientHttp2Stream | null = null;
 		let heartbeatTimer: NodeJS.Timeout | null = null;
 		let debugResponseLogPromise: Promise<RequestDebugResponseLog | undefined> | undefined;
+		const h2Completion = Promise.withResolvers<void>();
+		let resolveH2: (() => void) | undefined = h2Completion.resolve;
 
 		try {
 			const apiKey = options?.apiKey;
@@ -406,6 +408,7 @@ export const streamCursor: StreamFunction<"cursor-agent"> = (
 			} else {
 				h2Client = http2.connect(baseUrl);
 			}
+			h2Client.on("error", h2Completion.reject);
 
 			h2Request = h2Client.request(requestHeaders);
 
@@ -449,8 +452,6 @@ export const streamCursor: StreamFunction<"cursor-agent"> = (
 				conversationStateCache.set(conversationId, checkpoint);
 			};
 
-			let resolveH2: (() => void) | undefined;
-
 			h2Request.on("response", headers => {
 				debugResponseLogPromise = debugSession?.openResponseLog(
 					`HTTP/2 ${headers[":status"] ?? ""}`.trim(),
@@ -517,8 +518,6 @@ export const streamCursor: StreamFunction<"cursor-agent"> = (
 				}
 			});
 
-			h2Request.write(frameConnectMessage(requestBytes));
-
 			const sendHeartbeat = () => {
 				if (!h2Request || h2Request.closed) {
 					return;
@@ -530,57 +529,55 @@ export const streamCursor: StreamFunction<"cursor-agent"> = (
 				h2Request.write(frameConnectMessage(heartbeatBytes));
 			};
 
-			heartbeatTimer = setInterval(sendHeartbeat, 5000);
-
-			await new Promise<void>((resolve, reject) => {
-				resolveH2 = resolve;
+			const closeDebugLog = async (): Promise<void> => {
+				const log = await debugResponseLogPromise;
+				await log?.close();
+			};
 
-				const closeDebugLog = async (): Promise<void> => {
-					const log = await debugResponseLogPromise;
-					await log?.close();
-				};
+			h2Request.on("trailers", trailers => {
+				const status = trailers["grpc-status"];
+				const msg = trailers["grpc-message"];
+				if (status && status !== "0") {
+					void closeDebugLog().finally(() => {
+						h2Completion.reject(
+							new AIError.ProviderResponseError(
+								`gRPC error ${status}: ${decodeURIComponent(String(msg || ""))}`,
+								{ kind: "envelope" },
+							),
+						);
+					});
+				}
+			});
 
-				h2Request!.on("trailers", trailers => {
-					const status = trailers["grpc-status"];
-					const msg = trailers["grpc-message"];
-					if (status && status !== "0") {
-						void closeDebugLog().finally(() => {
-							reject(
-								new AIError.ProviderResponseError(
-									`gRPC error ${status}: ${decodeURIComponent(String(msg || ""))}`,
-									{ kind: "envelope" },
-								),
-							);
-						});
-					}
-				});
+			h2Request.on("end", () => {
+				resolveH2 = undefined;
+				void closeDebugLog()
+					.then(() => {
+						if (endStreamError) {
+							h2Completion.reject(endStreamError);
+							return;
+						}
+						h2Completion.resolve();
+					})
+					.catch(h2Completion.reject);
+			});
 
-				h2Request!.on("end", () => {
-					resolveH2 = undefined;
-					void closeDebugLog()
-						.then(() => {
-							if (endStreamError) {
-								reject(endStreamError);
-								return;
-							}
-							resolve();
-						})
-						.catch(reject);
-				});
+			h2Request.on("error", error => {
+				void closeDebugLog().finally(() => h2Completion.reject(error));
+			});
 
-				h2Request!.on("error", error => {
-					void closeDebugLog().finally(() => reject(error));
+			if (options?.signal) {
+				options.signal.addEventListener("abort", () => {
+					h2Request?.close();
+					void closeDebugLog().finally(() => {
+						h2Completion.reject(new AIError.AbortError());
+					});
 				});
+			}
 
-				if (options?.signal) {
-					options.signal.addEventListener("abort", () => {
-						h2Request?.close();
-						void closeDebugLog().finally(() => {
-							reject(new AIError.AbortError());
-						});
-					});
-				}
-			});
+			h2Request.write(frameConnectMessage(requestBytes));
+			heartbeatTimer = setInterval(sendHeartbeat, 5000);
+			await h2Completion.promise;
 
 			endCurrentTextBlock(output, stream, state);
 			endCurrentThinkingBlock(output, stream, state);

package/src/providers/ollama.ts

@@ -1,4 +1,4 @@
-import { fetchWithRetry, parseStreamingJson } from "@oh-my-pi/pi-utils";
+import { fetchWithRetry, parseStreamingJson, readJsonl } from "@oh-my-pi/pi-utils";
 import * as AIError from "../error";
 import { getEnvApiKey } from "../stream";
 import type {
@@ -347,36 +347,6 @@ async function captureHttpErrorResponse(response: Response): Promise<CapturedHtt
 	};
 }
 
-async function* iterateNdjson(stream: ReadableStream<Uint8Array>): AsyncGenerator<OllamaChatChunk> {
-	const reader = stream.getReader();
-	const decoder = new TextDecoder();
-	let buffer = "";
-	while (true) {
-		const { done, value } = await reader.read();
-		if (done) {
-			break;
-		}
-		buffer += decoder.decode(value, { stream: true });
-		while (true) {
-			const newlineIndex = buffer.indexOf("\n");
-			if (newlineIndex < 0) {
-				break;
-			}
-			const line = buffer.slice(0, newlineIndex).trim();
-			buffer = buffer.slice(newlineIndex + 1);
-			if (!line) {
-				continue;
-			}
-			yield JSON.parse(line) as OllamaChatChunk;
-		}
-	}
-	buffer += decoder.decode();
-	const tail = buffer.trim();
-	if (tail) {
-		yield JSON.parse(tail) as OllamaChatChunk;
-	}
-}
-
 function createEmptyOutput(model: Model<"ollama-chat">): AssistantMessage {
 	return {
 		role: "assistant",
@@ -622,7 +592,7 @@ const streamOllamaOnce = (
 				});
 			}
 			stream.push({ type: "start", partial: output });
-			for await (const chunk of iterateNdjson(response.body)) {
+			for await (const chunk of readJsonl<OllamaChatChunk>(response.body)) {
 				if (chunk.message?.thinking) {
 					suppressHealedThinking = true;
 					endActiveTextBlock();