@oh-my-pi/pi-ai 16.1.19 → 16.1.20

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [16.1.20] - 2026-06-25
+
+### Fixed
+
+- Fixed Ollama/Ollama Cloud native chat responses that finish with `done_reason: "length"` and no assistant content surfacing as a normal empty stop; they now become a context-window error instead of entering empty-stop retry recovery. ([#3464](https://github.com/can1357/oh-my-pi/issues/3464))
+- Fixed direct Anthropic Claude Sonnet/Haiku 4.5 requests serializing `output_config.effort`. The catalog classification (`packages/catalog/src/model-thinking.ts`) drove the `anthropic-budget-effort` branch in `buildParams`, which Anthropic's first-party Messages API rejects on Sonnet/Haiku 4.5 with HTTP 400 `This model does not support the effort parameter.` Sonnet/Haiku 4.5 now use plain `thinking.budget_tokens`; Opus 4.5 still emits `output_config.effort` because Anthropic supports it there. ([#3497](https://github.com/can1357/oh-my-pi/issues/3497))
+
 ## [16.1.19] - 2026-06-25
 
 ### Fixed

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "16.1.19",
+	"version": "16.1.20",
 	"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": "^2.12.0",
-		"@oh-my-pi/pi-catalog": "16.1.19",
-		"@oh-my-pi/pi-utils": "16.1.19",
-		"@oh-my-pi/pi-wire": "16.1.19",
+		"@oh-my-pi/pi-catalog": "16.1.20",
+		"@oh-my-pi/pi-utils": "16.1.20",
+		"@oh-my-pi/pi-wire": "16.1.20",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},

package/src/providers/ollama.ts

@@ -437,6 +437,17 @@ function mapDoneReason(doneReason: string | undefined, output: AssistantMessage)
 	return "stop";
 }
 
+const EMPTY_OLLAMA_LENGTH_COMPLETION_MESSAGE =
+	"Model returned no content: prompt filled the context window; raise Ollama num_ctx or shorten the prompt.";
+
+function hasVisibleAssistantContent(output: AssistantMessage): boolean {
+	return output.content.some(block => {
+		if (block.type === "text") return block.text.trim().length > 0;
+		if (block.type === "thinking") return block.thinking.trim().length > 0;
+		return block.type === "toolCall";
+	});
+}
+
 const OLLAMA_RETRY_DELAYS_MS = [2_000, 5_000, 10_000];
 
 export const streamOllama: StreamFunction<"ollama-chat"> = (
@@ -702,6 +713,10 @@ export const streamOllama: StreamFunction<"ollama-chat"> = (
 			}
 			endActiveThinkingBlock();
 			endActiveTextBlock();
+			if (output.stopReason === "length" && !hasVisibleAssistantContent(output)) {
+				output.stopReason = "error";
+				output.errorMessage = EMPTY_OLLAMA_LENGTH_COMPLETION_MESSAGE;
+			}
 			// Tool calls always mean "execute and continue" in the OpenAI/Ollama contract.
 			// If the turn produced tool-call blocks but reported a natural `stop`, promote
 			// to `toolUse` so the agent loop runs them (it gates execution on the stop
@@ -713,6 +728,11 @@ export const streamOllama: StreamFunction<"ollama-chat"> = (
 			if (firstTokenTime) {
 				output.ttft = firstTokenTime - startTime;
 			}
+			if (output.stopReason === "error") {
+				stream.push({ type: "error", reason: "error", error: output });
+				stream.end();
+				return;
+			}
 			const doneReason =
 				output.stopReason === "length" ? "length" : output.stopReason === "toolUse" ? "toolUse" : "stop";
 			stream.push({ type: "done", reason: doneReason, message: output });