@oh-my-pi/pi-coding-agent 13.12.5 → 13.12.6

package/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [13.12.6] - 2026-03-15
+### Changed
+
+- Updated llama.cpp model discovery to read context window from the `/props` endpoint's `default_generation_settings.n_ctx` field instead of using hardcoded 128000 default
+- Updated llama.cpp model discovery to detect vision capabilities from the `/props` endpoint's `modalities.vision` field instead of defaulting to text-only input
+- Changed llama.cpp `maxTokens` calculation to respect discovered context window limits, capping at 8192 or the server's context window, whichever is smaller
+
+### Fixed
+
+- Fixed llama.cpp auto-discovery to read context window and vision support from the native `/props` endpoint instead of relying on hardcoded defaults
+
 ## [13.12.5] - 2026-03-15
 
 ### Added

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-coding-agent",
-	"version": "13.12.5",
+	"version": "13.12.6",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -41,12 +41,12 @@
 	},
 	"dependencies": {
 		"@mozilla/readability": "^0.6",
-		"@oh-my-pi/omp-stats": "13.12.5",
-		"@oh-my-pi/pi-agent-core": "13.12.5",
-		"@oh-my-pi/pi-ai": "13.12.5",
-		"@oh-my-pi/pi-natives": "13.12.5",
-		"@oh-my-pi/pi-tui": "13.12.5",
-		"@oh-my-pi/pi-utils": "13.12.5",
+		"@oh-my-pi/omp-stats": "13.12.6",
+		"@oh-my-pi/pi-agent-core": "13.12.6",
+		"@oh-my-pi/pi-ai": "13.12.6",
+		"@oh-my-pi/pi-natives": "13.12.6",
+		"@oh-my-pi/pi-tui": "13.12.6",
+		"@oh-my-pi/pi-utils": "13.12.6",
 		"@sinclair/typebox": "^0.34",
 		"@xterm/headless": "^6.0",
 		"ajv": "^8.18",

package/src/config/model-registry.ts

@@ -372,6 +372,11 @@ type OllamaDiscoveredModelMetadata = {
 	contextWindow?: number;
 };
 
+type LlamaCppDiscoveredServerMetadata = {
+	contextWindow?: number;
+	input?: ("text" | "image")[];
+};
+
 /**
  * Resolve an API key config value to an actual key.
  * Checks environment variable first, then treats as literal.
@@ -416,6 +421,25 @@ function extractOllamaContextWindow(payload: Record<string, unknown>): number |
 	return match ? toPositiveNumberOrUndefined(match[1]) : undefined;
 }
 
+function extractLlamaCppContextWindow(payload: Record<string, unknown>): number | undefined {
+	const generationSettings = payload.default_generation_settings;
+	if (isRecord(generationSettings)) {
+		const contextWindow = toPositiveNumberOrUndefined(generationSettings.n_ctx);
+		if (contextWindow !== undefined) {
+			return contextWindow;
+		}
+	}
+	return toPositiveNumberOrUndefined(payload.n_ctx);
+}
+
+function extractLlamaCppInputCapabilities(payload: Record<string, unknown>): ("text" | "image")[] | undefined {
+	const modalities = payload.modalities;
+	if (!isRecord(modalities)) {
+		return undefined;
+	}
+	return modalities.vision === true ? ["text", "image"] : ["text"];
+}
+
 function extractGoogleOAuthToken(value: string | undefined): string | undefined {
 	if (!isAuthenticated(value)) return undefined;
 	try {
@@ -1225,6 +1249,32 @@ export class ModelRegistry {
 		return this.#applyProviderModelOverrides(providerConfig.provider, discovered);
 	}
 
+	async #discoverLlamaCppServerMetadata(
+		baseUrl: string,
+		headers: Record<string, string> | undefined,
+	): Promise<LlamaCppDiscoveredServerMetadata | null> {
+		const propsUrl = `${this.#toLlamaCppNativeBaseUrl(baseUrl)}/props`;
+		try {
+			const response = await fetch(propsUrl, {
+				headers,
+				signal: AbortSignal.timeout(150),
+			});
+			if (!response.ok) {
+				return null;
+			}
+			const payload = (await response.json()) as unknown;
+			if (!isRecord(payload)) {
+				return null;
+			}
+			return {
+				contextWindow: extractLlamaCppContextWindow(payload),
+				input: extractLlamaCppInputCapabilities(payload),
+			};
+		} catch {
+			return null;
+		}
+	}
+
 	async #discoverLlamaCppModels(providerConfig: DiscoveryProviderConfig): Promise<Model<Api>[]> {
 		const baseUrl = this.#normalizeLlamaCppBaseUrl(providerConfig.baseUrl);
 		const modelsUrl = `${baseUrl}/models`;
@@ -1235,10 +1285,13 @@ export class ModelRegistry {
 			headers.Authorization = `Bearer ${apiKey}`;
 		}
 
-		const response = await fetch(modelsUrl, {
-			headers,
-			signal: AbortSignal.timeout(250),
-		});
+		const [response, serverMetadata] = await Promise.all([
+			fetch(modelsUrl, {
+				headers,
+				signal: AbortSignal.timeout(250),
+			}),
+			this.#discoverLlamaCppServerMetadata(baseUrl, headers),
+		]);
 		if (!response.ok) {
 			throw new Error(`HTTP ${response.status} from ${modelsUrl}`);
 		}
@@ -1256,10 +1309,10 @@ export class ModelRegistry {
 					provider: providerConfig.provider,
 					baseUrl,
 					reasoning: false,
-					input: ["text"],
+					input: serverMetadata?.input ?? ["text"],
 					cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-					contextWindow: 128000,
-					maxTokens: 8192,
+					contextWindow: serverMetadata?.contextWindow ?? 128000,
+					maxTokens: Math.min(serverMetadata?.contextWindow ?? Number.POSITIVE_INFINITY, 8192),
 					headers,
 					compat: {
 						supportsStore: false,
@@ -1331,6 +1384,18 @@ export class ModelRegistry {
 		}
 	}
 
+	#toLlamaCppNativeBaseUrl(baseUrl: string): string {
+		try {
+			const parsed = new URL(baseUrl);
+			const trimmedPath = parsed.pathname.replace(/\/+$/g, "");
+			parsed.pathname = trimmedPath.endsWith("/v1") ? trimmedPath.slice(0, -3) || "/" : trimmedPath || "/";
+			const normalized = `${parsed.protocol}//${parsed.host}${parsed.pathname}`;
+			return normalized.endsWith("/") ? normalized.slice(0, -1) : normalized;
+		} catch {
+			return baseUrl.endsWith("/v1") ? baseUrl.slice(0, -3) : baseUrl;
+		}
+	}
+
 	#normalizeLmStudioBaseUrl(baseUrl?: string): string {
 		const defaultBaseUrl = "http://127.0.0.1:1234/v1";
 		const raw = baseUrl || defaultBaseUrl;