fix(ai): dropped nanogpt :tools route on deepseek

The :tools model suffix engages NanoGPT's server-side tool-call parser,
which 502s with code malformed_tool_call on complex DeepSeek payloads
(observed reliably on todo_write). The default route forwards
delta.content (including DSML envelope leaks) which our StreamMarkupHealing
already heals into a structured tool call.

The DSML allowlist still covers nanogpt and the parallel-index fix
remains; only the :tools suffix is removed.

Fixes #1488

packages/ai/CHANGELOG.md

@@ -10,6 +10,7 @@
 - Fixed GLM-5.x coding-plan OpenAI-compatible streams to use a longer default watchdog window, avoiding spurious `OpenAI completions stream stalled while waiting for the next event` errors during slow `glm-5.1` thinking/output phases. ([#1494](https://github.com/can1357/oh-my-pi/issues/1494))
 - Fixed `zhipu-coding-plan` model discovery and credential validation to use the dedicated GLM Coding Plan endpoint (`https://open.bigmodel.cn/api/coding/paas/v4`) instead of the general BigModel endpoint, preventing requests from consuming ordinary account balance. ([#1494](https://github.com/can1357/oh-my-pi/issues/1494))
 - Fixed DeepSeek tool calls failing on NanoGPT (e.g. `nanogpt/deepseek/deepseek-v4-pro` with reasoning enabled) by routing tool-bearing DeepSeek requests through NanoGPT's `:tools` model route and adding `nanogpt` to the DSML leak allowlist so streamed `<｜DSML｜tool_calls>...</｜DSML｜tool_calls>` envelopes are healed into structured tool calls instead of being passed through as visible text. ([#1488](https://github.com/can1357/oh-my-pi/issues/1488))
+- Fixed DeepSeek tool calls failing on NanoGPT (e.g. `nanogpt/deepseek/deepseek-v4-pro` with reasoning enabled) by adding `nanogpt` to the DSML leak allowlist so streamed `<｜DSML｜tool_calls>...</｜DSML｜tool_calls>` envelopes are healed into structured tool calls instead of being passed through as visible text. The `:tools` model suffix is no longer appended on NanoGPT; that route triggered NanoGPT's server-side tool-call parser and 502'd with `code: "malformed_tool_call"` on complex tool schemas (`todo_write`) — the default route forwards `delta.content` (including DSML envelopes) which is healed client-side. ([#1488](https://github.com/can1357/oh-my-pi/issues/1488))
 - Fixed OpenAI-compatible streamed parallel tool calls losing indexed argument deltas by tracking active tool-call blocks by the provider's `tool_calls[].index`; this keeps parallel NanoGPT `read` calls from merging or dropping their `path` arguments. ([#1488](https://github.com/can1357/oh-my-pi/issues/1488))
 
 ## [15.5.11] - 2026-05-29

packages/ai/src/providers/openai-completions.ts

@@ -92,24 +92,20 @@ function normalizeMistralToolId(id: string, isMistral: boolean): string {
 	}
 	return normalized;
 }
-
-// NanoGPT's default DeepSeek route can attempt server-side tool-call repair and
-// fail before streaming. `:tools` selects its documented tools-capable route.
-function shouldUseNanoGptToolsRoute(model: Model<"openai-completions">, context: Context): boolean {
-	return (
-		model.provider === "nanogpt" && !!context.tools?.length && /deepseek/i.test(model.id) && !model.id.includes(":")
-	);
-}
-
+// Direct DeepSeek model ids on NanoGPT are routed via the default tools-capable
+// path. We deliberately do NOT append `:tools` here: with `:tools`, NanoGPT
+// performs server-side tool-call parsing on the upstream DeepSeek stream and
+// 502s with `code: "malformed_tool_call"` on more complex tool schemas (issue
+// #1488). The default route forwards `delta.content` (including any DSML
+// envelope leaks) which `StreamMarkupHealing` heals into a structured call
+// client-side.
 function resolveOpenAICompletionsModelId(
 	model: Model<"openai-completions">,
-	context: Context,
 	options: OpenAICompletionsOptions | undefined,
 ): string {
 	if (model.provider === "firepass") return toFirepassWireModelId(model.id);
 	if (model.provider === "fireworks") return toFireworksWireModelId(model.id);
 	if (model.provider === "openrouter") return applyOpenRouterRoutingVariant(model.id, options?.openrouterVariant);
-	if (shouldUseNanoGptToolsRoute(model, context)) return `${model.id}:tools`;
 	return model.id;
 }
 
@@ -1140,7 +1136,7 @@ function buildParams(
 	// Note: Direct kimi-code provider is handled by the dedicated Kimi provider in kimi.ts.
 	const effectiveMaxTokens = options?.maxTokens ?? (isKimiModelId ? model.maxTokens : undefined);
 
-	const requestModelId = resolveOpenAICompletionsModelId(model, context, options);
+	const requestModelId = resolveOpenAICompletionsModelId(model, options);
 	const params: OpenAICompletionsParams = {
 		model: requestModelId,
 		messages,

packages/ai/test/stream-markup-healing.test.ts

@@ -659,7 +659,10 @@ describe("OpenAI completions provider DSML envelope healing", () => {
 			},
 		).result();
 
-		expect(payload?.model).toBe("deepseek/deepseek-v4-pro:tools");
+		// Issue #1488: `:tools` triggers NanoGPT's server-side tool-call parser
+		// which 502s on complex DeepSeek payloads. We route via the default
+		// path and rely on DSML healing instead.
+		expect(payload?.model).toBe("deepseek/deepseek-v4-pro");
 		expect(payload?.reasoning_effort).toBe("high");
 		expect(payload?.tools).toBeDefined();
 		const text = result.content