@oh-my-pi/pi-agent-core 18.13.2 → 18.13.3

package/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+## [18.13.3] - 2026-10-28
+
+### Changed
+
+- Updated Bun engine requirement to ≥1.4.4
+
+### Fixed
+
+- Fixed chat-completions context estimates counting past thinking that the request drops (no reasoning replay), which triggered compaction far below the threshold ([#15797](https://github.com/can1357/oh-my-pi/issues/15797), [#17231](https://github.com/can1357/oh-my-pi/pull/17231) by [@ghisguth](https://github.com/ghisguth))
+
 ## [18.13.2] - 2026-10-27
 
 ### Added

package/dist/types/tokenizer.d.ts

@@ -9,6 +9,7 @@ export declare function tokenizerEncodingForModel(model: Pick<Model, "tokenizer"
 type TokenizerModel = Pick<Model, "tokenizer"> & snapcompact.ShapeTarget & {
     compat?: Model["compat"];
     thinking?: Model["thinking"];
+    reasoning?: boolean;
 };
 /** Identity of the reader whose replayed reasoning a {@link Tokenizer} counts; equal keys count it alike. */
 export declare function reasoningReplayKey(model: TokenizerModel | null | undefined): string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-agent-core",
-	"version": "18.13.2",
+	"version": "18.13.3",
 	"description": "General-purpose agent with transport abstraction, state management, and attachment support",
 	"homepage": "https://omp.sh",
 	"author": {
@@ -38,22 +38,22 @@
 		"fmt": "oxfmt --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts'"
 	},
 	"dependencies": {
-		"@oh-my-pi/pi-ai": "18.13.2",
-		"@oh-my-pi/pi-catalog": "18.13.2",
-		"@oh-my-pi/pi-natives": "18.13.2",
-		"@oh-my-pi/pi-utils": "18.13.2",
-		"@oh-my-pi/pi-wire": "18.13.2",
-		"@oh-my-pi/snapcompact": "18.13.2",
+		"@oh-my-pi/pi-ai": "18.13.3",
+		"@oh-my-pi/pi-catalog": "18.13.3",
+		"@oh-my-pi/pi-natives": "18.13.3",
+		"@oh-my-pi/pi-utils": "18.13.3",
+		"@oh-my-pi/pi-wire": "18.13.3",
+		"@oh-my-pi/snapcompact": "18.13.3",
 		"@opentelemetry/api": "^1.9.1"
 	},
 	"devDependencies": {
-		"@oh-my-pi/omptype": "18.13.2",
+		"@oh-my-pi/omptype": "18.13.3",
 		"@opentelemetry/context-async-hooks": "^2.9.0",
 		"@opentelemetry/sdk-trace-base": "^2.9.0",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {
-		"bun": ">=1.3.14"
+		"bun": ">=1.4.4"
 	},
 	"files": [
 		"src",

package/src/tokenizer.ts

@@ -1,6 +1,7 @@
 import type { AssistantMessage, Model } from "@oh-my-pi/pi-ai";
 import { renderDemotedThinking } from "@oh-my-pi/pi-ai/dialect/demotion";
 import { ollamaChatDropsOwnThinking } from "@oh-my-pi/pi-ai/providers/ollama";
+import { completionsDropsOwnThinking, isLiteLLMThinkingSignature } from "@oh-my-pi/pi-ai/providers/openai-completions";
 import { isColdReplayableResponsesTurn } from "@oh-my-pi/pi-ai/providers/openai-shared";
 import {
 	isSameModelTurn,
@@ -41,7 +42,7 @@ export function tokenizerEncodingForModel(model: Pick<Model, "tokenizer"> | null
 }
 
 type TokenizerModel = Pick<Model, "tokenizer"> &
-	snapcompact.ShapeTarget & { compat?: Model["compat"]; thinking?: Model["thinking"] };
+	snapcompact.ShapeTarget & { compat?: Model["compat"]; thinking?: Model["thinking"]; reasoning?: boolean };
 
 /** The model an assistant turn's native reasoning (signed thinking, Responses history) is replayed to. */
 interface ReasoningReplayReader {
@@ -213,6 +214,7 @@ export class Tokenizer {
 	readonly frameBillingKey: string;
 	readonly #reasoningReader: ReasoningReplayReader | undefined;
 	readonly #dropsOllamaThinking: boolean;
+	readonly #dropsCompletionsThinking: boolean;
 	/** Claude's own API and prefix-bound hosts keep another model's signature from their deployment. */
 	readonly #readsDeploymentSignatures: boolean;
 	readonly reasoningReplayKey: string;
@@ -251,6 +253,7 @@ export class Tokenizer {
 		this.frameBillingKey = snapcompact.frameBillingKey(this.#frameBilling);
 		this.#reasoningReader = reasoningReplayReader(model);
 		this.#dropsOllamaThinking = ollamaChatDropsOwnThinking(model);
+		this.#dropsCompletionsThinking = completionsDropsOwnThinking(model);
 		const compat = model?.compat;
 		this.#readsDeploymentSignatures =
 			model?.api === "anthropic-messages" &&
@@ -424,6 +427,9 @@ export class Tokenizer {
 				const replayedReasoning = this.#replayedReasoningTokens(message);
 				const ollamaReader = this.#reasoningReader?.api === "ollama-chat" ? this.#reasoningReader : undefined;
 				const ownOllamaTurn = ollamaReader !== undefined && isSameModelTurn(message, ollamaReader);
+				const completionsReader = this.#dropsCompletionsThinking ? this.#reasoningReader : undefined;
+				const ownDroppedCompletionsTurn =
+					completionsReader !== undefined && isSameModelTurn(message, completionsReader);
 				if (typeof replayedReasoning === "number") extra += replayedReasoning;
 				const signaturesPriced = this.thinkingReplayTokens === "signature";
 				// The replay's signature rule (`transformMessages`): an aborted or
@@ -457,6 +463,8 @@ export class Tokenizer {
 							continue;
 						}
 						if (this.#dropsOllamaThinking && ownOllamaTurn) continue;
+						// Chat-completions drops it unless LiteLLM replays it in `thinking_blocks`.
+						if (ownDroppedCompletionsTurn && !isLiteLLMThinkingSignature(block.thinkingSignature)) continue;
 						fragments.push(block.thinking);
 						// Providers charge for the opaque signature/reasoning payload that
 						// rides alongside the thinking text (OpenAI Responses encrypted
@@ -578,6 +586,7 @@ export class Tokenizer {
 			reasoningReplayKey(model) === this.reasoningReplayKey &&
 			resolveThinkingReplayTokens(model ?? {}) === this.thinkingReplayTokens &&
 			ollamaChatDropsOwnThinking(model) === this.#dropsOllamaThinking &&
+			completionsDropsOwnThinking(model) === this.#dropsCompletionsThinking &&
 			(options?.nativeHistoryReplay ?? true) === this.nativeHistoryReplay
 		);
 	}