@oh-my-pi/pi-ai 16.1.2 → 16.1.3

package/CHANGELOG.md

@@ -2,6 +2,22 @@
 
 ## [Unreleased]
 
+## [16.1.3] - 2026-06-19
+
+### Added
+
+- Added regression test pinning that `openai-completions` emits a `thinking` block for `reasoning_content` deltas even when `delta.content` is explicitly JSON `null` (the DeepSeek-format dual-key pattern used by custom GLM/Qwen reasoning providers). See [#2996](https://github.com/can1357/oh-my-pi/issues/2996).
+
+### Changed
+
+- Improved the thinking loop guard to treat assistant text loops as retryable errors
+- Refined text normalization logic to reduce false positives in the thinking loop detector
+
+### Fixed
+
+- Fixed Ollama chat requests sending image payloads to text-only models. Image blocks are now omitted and replaced with the standard non-vision placeholder for models without vision support, while vision-capable Ollama models continue to receive images. ([#3009](https://github.com/can1357/oh-my-pi/pull/3009) by [@serverinspector](https://github.com/serverinspector))
+- Fixed `SqliteAuthCredentialStore.close()` leaking one-off prepared statements created by inline `this.#db.prepare()` calls in `#authCredentialsTableExists`, `#readAuthSchemaVersion`, `#inferAuthSchemaVersion`, `#migrateAuthSchemaV0ToV1`, `#backfillCredentialIdentityKeys`, and `updateAuthCredential`. Each statement is now wrapped in `try/finally` with `stmt.finalize()`, and the `close()` method finalizes `#insertUsageCostStmt` and `#listUsageCostsStmt` which were previously missed. This caused EBUSY on Windows when tests tried to delete temp dirs containing open SQLite handles.
+
 ## [16.1.2] - 2026-06-19
 
 ### Added
@@ -3934,4 +3950,4 @@ _Dedicated to Peter's shoulder ([@steipete](https://twitter.com/steipete))_
 
 ## [0.9.4] - 2025-11-26
 
-Initial release with multi-provider LLM support.
+Initial release with multi-provider LLM support.

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "16.1.2",
+	"version": "16.1.3",
 	"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": "16.1.2",
-		"@oh-my-pi/pi-utils": "16.1.2",
-		"@oh-my-pi/pi-wire": "16.1.2",
+		"@oh-my-pi/pi-catalog": "16.1.3",
+		"@oh-my-pi/pi-utils": "16.1.3",
+		"@oh-my-pi/pi-wire": "16.1.3",
 		"arktype": "^2.2.0",
 		"partial-json": "^0.1.7",
 		"zod": "^4"

package/src/auth-storage.ts

@@ -4742,25 +4742,47 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 	}
 
 	#authCredentialsTableExists(): boolean {
-		const row = this.#db
-			.prepare("SELECT 1 AS present FROM sqlite_master WHERE type = 'table' AND name = 'auth_credentials'")
-			.get() as { present?: number } | undefined;
-		return row?.present === 1;
+		const stmt = this.#db.prepare(
+			"SELECT 1 AS present FROM sqlite_master WHERE type = 'table' AND name = 'auth_credentials'",
+		);
+		try {
+			const row = stmt.get() as { present?: number } | undefined;
+			return row?.present === 1;
+		} finally {
+			stmt.finalize();
+		}
 	}
 
 	#readAuthSchemaVersion(): number | null {
-		const row = this.#db.prepare("SELECT version FROM auth_schema_version WHERE id = 1").get() as
-			| { version?: number }
-			| undefined;
-		return typeof row?.version === "number" ? row.version : null;
+		const stmt = this.#db.prepare("SELECT version FROM auth_schema_version WHERE id = 1");
+		try {
+			const row = stmt.get() as { version?: number } | undefined;
+			return typeof row?.version === "number" ? row.version : null;
+		} finally {
+			stmt.finalize();
+		}
 	}
 
 	#writeAuthSchemaVersion(version: number): void {
-		this.#db.prepare("INSERT OR REPLACE INTO auth_schema_version(id, version) VALUES (1, ?)").run(version);
+		const stmt = this.#db.prepare("INSERT OR REPLACE INTO auth_schema_version(id, version) VALUES (1, ?)");
+		try {
+			stmt.run(version);
+		} finally {
+			stmt.finalize();
+		}
 	}
 
 	#inferAuthSchemaVersion(): number {
-		const cols = this.#db.prepare("PRAGMA table_info(auth_credentials)").all() as Array<{ name?: string }>;
+		const stmt = this.#db.prepare("PRAGMA table_info(auth_credentials)");
+		try {
+			const cols = stmt.all() as Array<{ name?: string }>;
+			return this.#inferAuthSchemaVersionFromColumns(cols);
+		} finally {
+			stmt.finalize();
+		}
+	}
+
+	#inferAuthSchemaVersionFromColumns(cols: Array<{ name?: string }>): number {
 		const hasDisabledCause = cols.some(column => column.name === "disabled_cause");
 		const hasIdentityKey = cols.some(column => column.name === "identity_key");
 		const hasAccountId = cols.some(column => column.name === "account_id");
@@ -4808,8 +4830,14 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 
 	#migrateAuthSchemaV0ToV1(): void {
 		const migrate = this.#db.transaction(() => {
-			const v0Cols = this.#db.prepare("PRAGMA table_info(auth_credentials)").all() as Array<{ name?: string }>;
-			const hasDisabled = v0Cols.some(col => col.name === "disabled");
+			const stmt = this.#db.prepare("PRAGMA table_info(auth_credentials)");
+			let hasDisabled = false;
+			try {
+				const v0Cols = stmt.all() as Array<{ name?: string }>;
+				hasDisabled = v0Cols.some(col => col.name === "disabled");
+			} finally {
+				stmt.finalize();
+			}
 
 			this.#db.run("ALTER TABLE auth_credentials RENAME TO auth_credentials_v0");
 			this.#db.run(`
@@ -4885,21 +4913,29 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 	}
 
 	#backfillCredentialIdentityKeys(): void {
-		const rows = this.#db
-			.prepare(
-				"SELECT id, provider, credential_type, data, disabled_cause, identity_key FROM auth_credentials WHERE identity_key IS NULL ORDER BY id ASC",
-			)
-			.all() as AuthRow[];
+		const selectRowsStmt = this.#db.prepare(
+			"SELECT id, provider, credential_type, data, disabled_cause, identity_key FROM auth_credentials WHERE identity_key IS NULL ORDER BY id ASC",
+		);
+		let rows: AuthRow[];
+		try {
+			rows = selectRowsStmt.all() as AuthRow[];
+		} finally {
+			selectRowsStmt.finalize();
+		}
 		if (rows.length === 0) return;
 
 		let updateIdentity: Statement | null = null;
-		for (const row of rows) {
-			const identityKey = resolveRowCredentialIdentityKey(row.provider, row);
-			// Rows whose identity cannot be derived stay NULL; writing NULL over
-			// NULL would just burn a write transaction on every boot.
-			if (identityKey === null) continue;
-			updateIdentity ??= this.#db.prepare("UPDATE auth_credentials SET identity_key = ? WHERE id = ?");
-			updateIdentity.run(identityKey, row.id);
+		try {
+			for (const row of rows) {
+				const identityKey = resolveRowCredentialIdentityKey(row.provider, row);
+				// Rows whose identity cannot be derived stay NULL; writing NULL over
+				// NULL would just burn a write transaction on every boot.
+				if (identityKey === null) continue;
+				updateIdentity ??= this.#db.prepare("UPDATE auth_credentials SET identity_key = ? WHERE id = ?");
+				updateIdentity.run(identityKey, row.id);
+			}
+		} finally {
+			updateIdentity?.finalize();
 		}
 	}
 
@@ -5063,9 +5099,13 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 
 	updateAuthCredential(id: number, credential: AuthCredential): void {
 		try {
-			const providerRow = this.#db.prepare("SELECT provider FROM auth_credentials WHERE id = ?").get(id) as
-				| { provider?: string }
-				| undefined;
+			const providerStmt = this.#db.prepare("SELECT provider FROM auth_credentials WHERE id = ?");
+			let providerRow: { provider?: string } | undefined;
+			try {
+				providerRow = providerStmt.get(id) as { provider?: string } | undefined;
+			} finally {
+				providerStmt.finalize();
+			}
 			const provider = providerRow?.provider ?? "";
 			const serialized = serializeCredential(provider, credential);
 			if (!serialized) return;
@@ -5334,6 +5374,8 @@ export class SqliteAuthCredentialStore implements AuthCredentialStore {
 		this.#lastUsageHistoryStmt.finalize();
 		this.#listUsageHistoryStmt.finalize();
 		this.#updateUsageHistoryStmt.finalize();
+		this.#insertUsageCostStmt.finalize();
+		this.#listUsageCostsStmt.finalize();
 		this.#db.close();
 	}
 }

package/src/providers/ollama.ts

@@ -5,15 +5,14 @@ import type {
 	Api,
 	AssistantMessage,
 	Context,
-	DeveloperMessage,
+	ImageContent,
 	Message,
 	Model,
 	StreamFunction,
 	StreamOptions,
+	TextContent,
 	Tool,
 	ToolChoice,
-	ToolResultMessage,
-	UserMessage,
 } from "../types";
 import { normalizeSystemPrompts } from "../utils";
 import { AssistantMessageEventStream } from "../utils/event-stream";
@@ -32,6 +31,7 @@ import {
 	type StreamMarkupHealingEvent,
 } from "../utils/stream-markup-healing";
 import { transformMessages } from "./transform-messages";
+import { joinTextWithImagePlaceholder, partitionVisionContent } from "./vision-guard";
 
 /** Non-2xx response from the Ollama `/api/chat` endpoint. */
 export class OllamaApiError extends ProviderHttpError {
@@ -174,40 +174,35 @@ function selectToolsForToolChoice(tools: Tool[] | undefined, toolChoice: ToolCho
 	return [];
 }
 
-function toPlainContent(content: string | Array<{ type: "text" | "image"; text?: string; data?: string }>): {
+function toPlainContent(
+	content: string | ReadonlyArray<TextContent | ImageContent>,
+	supportsImages: boolean,
+): {
 	content: string;
 	images?: string[];
 } {
 	if (typeof content === "string") {
 		return { content };
 	}
-	const textParts: string[] = [];
-	const images: string[] = [];
-	for (const block of content) {
-		if (block.type === "text" && typeof block.text === "string") {
-			textParts.push(block.text);
-		}
-		if (block.type === "image" && typeof block.data === "string") {
-			images.push(block.data);
-		}
-	}
+	const { textBlocks, imageBlocks, omittedImages } = partitionVisionContent(content, supportsImages);
+	const text = textBlocks.map(block => block.text).join("\n");
 	return {
-		content: textParts.join("\n"),
-		...(images.length > 0 ? { images } : {}),
+		content: joinTextWithImagePlaceholder(text, omittedImages),
+		...(imageBlocks.length > 0 ? { images: imageBlocks.map(block => block.data) } : {}),
 	};
 }
 
-function convertMessage(message: Message): OllamaMessage {
+function convertMessage(message: Message, supportsImages: boolean): OllamaMessage {
 	if (message.role === "user") {
-		const converted = toPlainContent(message.content as UserMessage["content"]);
+		const converted = toPlainContent(message.content, supportsImages);
 		return { role: "user", ...converted };
 	}
 	if (message.role === "developer") {
-		const converted = toPlainContent(message.content as DeveloperMessage["content"]);
+		const converted = toPlainContent(message.content, supportsImages);
 		return { role: "system", ...converted };
 	}
 	if (message.role === "toolResult") {
-		const converted = toPlainContent(message.content as ToolResultMessage["content"]);
+		const converted = toPlainContent(message.content, supportsImages);
 		return {
 			role: "tool",
 			tool_name: message.toolName,
@@ -259,8 +254,9 @@ function convertMessages(model: Model<"ollama-chat">, context: Context): OllamaM
 	}
 	messages.push(...context.messages);
 	const isCloud = model.provider === "ollama-cloud";
+	const supportsImages = model.input.includes("image");
 	return transformMessages(messages, model).map(msg => {
-		const converted = convertMessage(msg);
+		const converted = convertMessage(msg, supportsImages);
 		// Ollama cloud rejects requests when assistant history messages contain the `thinking`
 		// field — it's valid in model responses but not accepted as a history input. Strip it
 		// to prevent HTTP 400 errors. Local Ollama instances are unaffected.

package/src/utils/thinking-loop.ts

@@ -25,11 +25,11 @@
  *    13.5k non-loop thinking blocks (zero false positives; hardest negative
  *    scored 3 against the trigger of 4).
  *
- * Scope is deliberately narrow: **thinking only**. Answer text is left
- * untouched so the guard can never discard already-streamed visible output. The
- * guard is gated to Gemini models and wraps the provider stream, so it works
- * across every Gemini transport (OpenRouter `openai-completions`, direct
- * `google-generative-ai` / `google-gemini-cli`, Vertex). Disable with
+ * Scope is narrow: guarded Gemini/DeepSeek streams before any tool call. Native
+ * thinking is checked first; assistant text can also be checked for providers
+ * that surface reasoning as visible prose. On a hit, the failed turn is emitted
+ * as an empty retryable stream-stall error so the session drops and re-samples
+ * it instead of committing the runaway transcript. Disable with
  * `PI_NO_THINKING_LOOP_GUARD=1`.
  */
 import { logger } from "@oh-my-pi/pi-utils";
@@ -208,7 +208,6 @@ export function guardThinkingLoopStream(
 	void (async () => {
 		let thinkingArmed = true;
 		let textArmed = checkAssistantContent;
-		let accumulatedText = "";
 		try {
 			for await (const event of inner) {
 				let detail: string | null = null;
@@ -221,7 +220,6 @@ export function guardThinkingLoopStream(
 					thinkingArmed = false;
 					if (textArmed && event.type === "text_delta") {
 						detail = textDetector.push(event.delta);
-						accumulatedText += event.delta;
 					}
 				} else if (event.type === "toolcall_start" || event.type === "toolcall_delta") {
 					thinkingArmed = false;
@@ -244,7 +242,7 @@ export function guardThinkingLoopStream(
 					outer.push({
 						type: "error",
 						reason: "error",
-						error: buildThinkingLoopError(model, detail, accumulatedText),
+						error: buildThinkingLoopError(model, detail),
 					});
 					return;
 				}
@@ -289,13 +287,13 @@ export function withGeminiThinkingLoopGuard<
 	return guardThinkingLoopStream(dispatch(merged), model, controller, options);
 }
 
-function buildThinkingLoopError(model: Model<Api>, detail: string, accumulatedText?: string): AssistantMessage {
-	const hasText = Boolean(accumulatedText);
+function buildThinkingLoopError(model: Model<Api>, detail: string): AssistantMessage {
 	return {
 		role: "assistant",
-		// Empty content is load-bearing: a contentful error stop is replay-unsafe
-		// and would NOT be auto-retried by the session.
-		content: hasText ? [{ type: "text", text: accumulatedText! }] : [],
+		// Empty content is load-bearing: loop-guard output is replay garbage, even
+		// when it arrived as assistant text instead of native thinking. Keeping it
+		// would persist the failed attempt before AgentSession retries.
+		content: [],
 		api: model.api,
 		provider: model.provider,
 		model: model.id,
@@ -310,7 +308,7 @@ function buildThinkingLoopError(model: Model<Api>, detail: string, accumulatedTe
 		stopReason: "error",
 		// "stream stall" makes the transport/session retry classifiers treat this
 		// as a transient (retryable) failure with no bespoke rule.
-		errorMessage: `${THINKING_LOOP_ERROR_MARKER}: the model repeated near-identical content (${detail}).${hasText ? " Non-retryable because output was already streamed." : " Treating as a stream stall and retrying."}`,
+		errorMessage: `${THINKING_LOOP_ERROR_MARKER}: the model repeated near-identical content (${detail}). Treating as a stream stall and retrying.`,
 		timestamp: Date.now(),
 	};
 }
@@ -345,14 +343,15 @@ function detectVerbatimRepetition(text: string): [unit: string, count: number] |
 	return null;
 }
 
-/** Lowercase, drop code spans / paths / digits, keep only letter words. */
+/** Lowercase and tokenize prose plus code/path payloads, dropping pure numbers. */
 function normalizeSegment(segment: string): string {
 	return segment
 		.toLowerCase()
-		.replace(/`[^`]*`/g, " ")
-		.replace(/\/[^\s`]+/g, " ")
-		.replace(/\d+/g, " ")
-		.replace(/[^a-z]+/g, " ")
+		.replace(/`([^`]*)`/g, " $1 ")
+		.replace(/[^a-z0-9]+/g, " ")
+		.split(/\s+/)
+		.filter(token => /[a-z]/.test(token))
+		.join(" ")
 		.trim();
 }