@f5-sales-demo/xcsh 21.44.1 → 21.45.0

package/CHANGELOG.md

@@ -6,6 +6,9 @@
 
 ### Added
 
+- Refreshed generated internal LiteLLM configuration with seven current GPT-6/Claude models,
+  route-accurate Responses and native Messages transports, hidden legacy session resolution, and an
+  independent maximum-context setting ([#4279](https://github.com/f5-sales-demo/xcsh/issues/4279)).
 - Added current ChatGPT subscription roles for GPT-6 Luna/Sol, Claude Opus 5.5 selection, and an
   opt-in Providers setting for the GPT-6 1.05M-token context window. The default remains 272K to
   make long-context pricing explicit; LiteLLM retains its existing catalog

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@f5-sales-demo/xcsh",
-	"version": "21.44.1",
+	"version": "21.45.0",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"homepage": "https://github.com/f5-sales-demo/xcsh",
 	"author": "Can Boluk",
@@ -64,13 +64,13 @@
 	},
 	"dependencies": {
 		"@agentclientprotocol/sdk": "1.4.0",
-		"@f5-sales-demo/pi-agent-core": "21.44.1",
-		"@f5-sales-demo/pi-ai": "21.44.1",
-		"@f5-sales-demo/pi-natives": "21.44.1",
-		"@f5-sales-demo/pi-resource-management": "21.44.1",
-		"@f5-sales-demo/pi-tui": "21.44.1",
-		"@f5-sales-demo/pi-utils": "21.44.1",
-		"@f5-sales-demo/xcsh-stats": "21.44.1",
+		"@f5-sales-demo/pi-agent-core": "21.45.0",
+		"@f5-sales-demo/pi-ai": "21.45.0",
+		"@f5-sales-demo/pi-natives": "21.45.0",
+		"@f5-sales-demo/pi-resource-management": "21.45.0",
+		"@f5-sales-demo/pi-tui": "21.45.0",
+		"@f5-sales-demo/pi-utils": "21.45.0",
+		"@f5-sales-demo/xcsh-stats": "21.45.0",
 		"@mozilla/readability": "^0.6",
 		"@sinclair/typebox": "0.34.52",
 		"@tobilu/qmd": "2.8.3",

package/src/commit/agentic/index.ts

@@ -31,6 +31,8 @@ export async function runAgenticCommit(args: CommitCommandArgs): Promise<void> {
 	process.stdout.write("● Resolving model...\n");
 	const modelRegistry = new ModelRegistry(authStorage, undefined, {
 		getProviderOrder: () => settings.get("modelProviderOrder"),
+		getOpenAICodexMaxContext: () => settings.get("providers.openaiCodexMaxContext"),
+		getLiteLLMMaxContext: () => settings.get("providers.litellmMaxContext"),
 	});
 	await modelRegistry.refresh();
 	const stagedFilesPromise = (async () => {

package/src/commit/pipeline.ts

@@ -44,6 +44,8 @@ async function runLegacyCommitCommand(args: CommitCommandArgs): Promise<void> {
 	const authStorage = await discoverAuthStorage();
 	const modelRegistry = new ModelRegistry(authStorage, undefined, {
 		getProviderOrder: () => settings.get("modelProviderOrder"),
+		getOpenAICodexMaxContext: () => settings.get("providers.openaiCodexMaxContext"),
+		getLiteLLMMaxContext: () => settings.get("providers.litellmMaxContext"),
 	});
 	await modelRegistry.refresh();
 

package/src/config/auto-config.ts

@@ -20,7 +20,7 @@ import { hardenAgentConfigFileSync, writeAgentConfigFileSync } from "./agent-con
 import { DEFAULT_MODEL_ROLE } from "./settings-schema";
 
 /** Current config schema version. Bump when the generated format changes. */
-export const CURRENT_CONFIG_VERSION = 8;
+export const CURRENT_CONFIG_VERSION = 9;
 const LITELLM_CONFIG_DIR_MODE = 0o700;
 const LITELLM_MODELS_FILE_MODE = 0o600;
 const GENERATED_LITELLM_MARKER = "# Auto-generated by xcsh for LiteLLM proxy";
@@ -64,7 +64,7 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 	// LiteLLM's OpenAI-compatible chat route can live behind a custom API prefix
 	// (for example /api/v1). Astra with tools and reasoning requires the
 	// Responses API, which LiteLLM exposes on its dedicated OpenAI route.
-	const astraResponsesBaseUrl = `${baseUrl.replace(/\/+$/, "")}/openai/v1`;
+	const responsesBaseUrl = `${baseUrl.replace(/\/+$/, "")}/openai/v1`;
 	// When a literal API key is provided, double-quote it to handle YAML-significant
 	// characters (: ! # { } [ ] ' " etc.). Escape backslashes and double quotes first.
 	const apiKeyValue = options?.apiKeyLiteral
@@ -81,14 +81,35 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"  anthropic:",
 		`    baseUrl: "${baseUrl}/anthropic"`,
 		`    apiKey: ${apiKeyValue}`,
+		"    api: anthropic-messages",
 		"    modelAllowlist:",
-		"      - claude-opus-5",
-		"      - claude-sonnet-5",
 		"      - claude-haiku-4-5",
+		"      - claude-sonnet-5",
+		"      - claude-opus-5-5",
+		"      - claude-opus-5",
 		"    picker:",
 		"      groupId: litellm",
 		"      groupLabel: LiteLLM",
 		"      sectionLabel: Anthropic",
+		"    modelOverrides:",
+		"      claude-haiku-4-5:",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        contextWindow: 200000",
+		"        maxTokens: 64000",
+		"        compat: { supportsToolChoice: false }",
+		"      claude-sonnet-5:",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        contextWindow: 1000000",
+		"        maxTokens: 128000",
+		"        compat: { supportsToolChoice: false }",
+		"      claude-opus-5-5:",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        contextWindow: 1000000",
+		"        maxTokens: 128000",
+		"        compat: { supportsToolChoice: false }",
+		"      claude-opus-5:",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        compat: { supportsToolChoice: false }",
 		"  litellm:",
 		`    baseUrl: "${baseUrl}${apiBase}"`,
 		`    apiKey: ${apiKeyValue}`,
@@ -96,25 +117,105 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"    discovery:",
 		"      type: openai-compat",
 		"    modelAllowlist:",
-		"      - gpt-6-astra",
-		"      - gpt-5.6-sol",
+		"      - gpt-6-luna",
 		"      - gpt-5.6-terra",
+		"      - gpt-6-sol",
+		"      - gpt-6-astra",
 		"      - gpt-5.6-luna",
+		"      - gpt-5.6-sol",
 		"    models:",
+		"      - id: gpt-6-luna",
+		"        name: GPT-6 Luna",
+		"        api: openai-responses",
+		`        baseUrl: "${responsesBaseUrl}"`,
+		"      - id: gpt-5.6-terra",
+		"        name: GPT-5.6 Terra",
+		"        api: openai-completions",
+		"      - id: gpt-6-sol",
+		"        name: GPT-6 Sol",
+		"        api: openai-responses",
+		`        baseUrl: "${responsesBaseUrl}"`,
 		"      - id: gpt-6-astra",
 		"        name: GPT-6 Astra",
 		"        api: openai-responses",
-		`        baseUrl: "${astraResponsesBaseUrl}"`,
+		`        baseUrl: "${responsesBaseUrl}"`,
+		"      - id: gpt-5.6-luna",
+		"        name: GPT-5.6 Luna",
+		"        api: openai-completions",
+		"      - id: gpt-5.6-sol",
+		"        name: GPT-5.6 Sol",
+		"        api: openai-completions",
 		"    picker:",
 		"      groupId: litellm",
 		"      groupLabel: LiteLLM",
 		"      sectionLabel: OpenAI",
 		"    modelOverrides:",
+		"      gpt-6-luna:",
+		"        reasoning: true",
+		"        input:",
+		"          - text",
+		"          - image",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        thinking:",
+		"          mode: effort",
+		"          defaultLevel: medium",
+		"          supportedLevels:",
+		"            - { effort: none, description: No reasoning }",
+		"            - { effort: low, description: Light reasoning }",
+		"            - { effort: medium, description: Balanced reasoning }",
+		"            - { effort: high, description: Deep reasoning }",
+		"            - { effort: xhigh, description: Very deep reasoning }",
+		"            - { effort: max, description: Maximum reasoning }",
+		"        contextWindow: 1050000",
+		"        maxTokens: 128000",
+		"        compat:",
+		"          supportsTemperature: false",
+		"      gpt-5.6-terra:",
+		"        reasoning: true",
+		"        input:",
+		"          - text",
+		"          - image",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        thinking:",
+		"          mode: effort",
+		"          defaultLevel: medium",
+		"          supportedLevels:",
+		"            - { effort: none, description: No reasoning }",
+		"            - { effort: low, description: Light reasoning }",
+		"            - { effort: medium, description: Balanced reasoning }",
+		"            - { effort: high, description: Deep reasoning }",
+		"            - { effort: xhigh, description: Very deep reasoning }",
+		"            - { effort: max, description: Maximum reasoning }",
+		"        contextWindow: 1050000",
+		"        maxTokens: 128000",
+		"        compat:",
+		"          supportsTemperature: false",
+		"      gpt-6-sol:",
+		"        reasoning: true",
+		"        input:",
+		"          - text",
+		"          - image",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        thinking:",
+		"          mode: effort",
+		"          defaultLevel: medium",
+		"          supportedLevels:",
+		"            - { effort: none, description: No reasoning }",
+		"            - { effort: low, description: Light reasoning }",
+		"            - { effort: medium, description: Balanced reasoning }",
+		"            - { effort: high, description: Deep reasoning }",
+		"            - { effort: xhigh, description: Very deep reasoning }",
+		"            - { effort: max, description: Maximum reasoning }",
+		"        contextWindow: 1050000",
+		"        maxTokens: 128000",
+		"        compat:",
+		"          supportsTemperature: false",
 		"      gpt-6-astra:",
 		"        reasoning: true",
 		"        input:",
 		"          - text",
 		"          - image",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
 		"        thinking:",
 		"          mode: effort",
 		"          defaultLevel: medium",
@@ -133,6 +234,7 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"        input:",
 		"          - text",
 		"          - image",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
 		"        thinking:",
 		"          mode: effort",
 		"          defaultLevel: medium",
@@ -147,10 +249,9 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"        maxTokens: 128000",
 		"        compat:",
 		"          supportsTemperature: false",
-		"      gpt-5.6-terra:",
-		"        reasoning: true",
 		"      gpt-5.6-luna:",
 		"        reasoning: true",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
 	];
 
 	lines.push("");

package/src/config/model-catalog.ts

@@ -12,6 +12,7 @@ function compareVersions(left: string, right: string): number {
 
 /** Keep historical catalogs available while presenting only current user-facing families. */
 const DEFAULT_OPENAI_CODEX_MODEL_IDS = new Set(["gpt-5.6-terra", "gpt-6-astra", "gpt-6-luna", "gpt-6-sol"]);
+const CURRENT_LITELLM_OPENAI_MODEL_IDS = new Set(["gpt-6-luna", "gpt-5.6-terra", "gpt-6-sol", "gpt-6-astra"]);
 
 /** Shared catalog boundary for the TUI and attached human interfaces. */
 export function filterCurrentBrowserModels(models: readonly Model[]): Model[] {
@@ -40,6 +41,7 @@ export function filterCurrentBrowserModels(models: readonly Model[]): Model[] {
 		if (gpt?.[1]) {
 			if (compareVersions(gpt[1], "5.6") < 0) return false;
 			if (model.provider === "openai-codex") return DEFAULT_OPENAI_CODEX_MODEL_IDS.has(model.id.toLowerCase());
+			if (model.provider === "litellm") return CURRENT_LITELLM_OPENAI_MODEL_IDS.has(model.id.toLowerCase());
 		}
 		const claude = model.id.match(/^claude-(opus|sonnet|haiku)-(\d+)(?:[.-](\d+))?(?:-|$)/i);
 		if (claude?.[1] && claude[2]) {

package/src/config/model-registry.ts

@@ -53,6 +53,9 @@ export const kNoAuth = NO_AUTH_API_KEY;
 const OPENAI_CODEX_STANDARD_CONTEXT_WINDOW = 272_000;
 const OPENAI_CODEX_MAX_CONTEXT_WINDOW = 1_050_000;
 const OPENAI_CODEX_MAX_CONTEXT_MODEL_IDS = new Set(["gpt-6-luna", "gpt-6-sol"]);
+const LITELLM_STANDARD_CONTEXT_WINDOW = 272_000;
+const LITELLM_MAX_CONTEXT_WINDOW = 1_050_000;
+const LITELLM_MAX_CONTEXT_MODEL_IDS = new Set(["gpt-6-luna", "gpt-5.6-terra", "gpt-6-sol", "gpt-6-astra"]);
 
 export function isAuthenticated(apiKey: string | undefined | null): apiKey is string {
 	return Boolean(apiKey) && apiKey !== kNoAuth;
@@ -876,6 +879,7 @@ export class ModelRegistry {
 	#hasProbed = false;
 	#refreshQueue: Promise<void> = Promise.resolve();
 	#openAICodexMaxContext = false;
+	#litellmMaxContext = false;
 
 	#enqueueRefresh(operation: () => Promise<void>): Promise<void> {
 		const pending = this.#refreshQueue.then(operation);
@@ -898,6 +902,7 @@ export class ModelRegistry {
 		private readonly options: {
 			getProviderOrder?: () => readonly string[];
 			getOpenAICodexMaxContext?: () => boolean;
+			getLiteLLMMaxContext?: () => boolean;
 		} = {},
 	) {
 		this.#modelsConfigFile = ModelsConfigFile.relocate(modelsPath);
@@ -1051,6 +1056,7 @@ export class ModelRegistry {
 
 	#loadModels() {
 		this.#openAICodexMaxContext = this.options.getOpenAICodexMaxContext?.() ?? this.#openAICodexMaxContext;
+		this.#litellmMaxContext = this.options.getLiteLLMMaxContext?.() ?? this.#litellmMaxContext;
 		// Load custom models from models.json first (to know which providers to override)
 		const {
 			models: customModels = [],
@@ -1085,6 +1091,7 @@ export class ModelRegistry {
 
 		this.#models = this.#applyProviderModelAllowlists(this.#applyModelOverrides(combined, this.#modelOverrides));
 		this.#applyOpenAICodexContextWindow();
+		this.#applyLiteLLMContextWindow();
 		this.#rebuildCanonicalIndex();
 	}
 
@@ -1106,6 +1113,22 @@ export class ModelRegistry {
 		this.#rebuildCanonicalIndex();
 	}
 
+	#applyLiteLLMContextWindow(): void {
+		const contextWindow = this.#litellmMaxContext ? LITELLM_MAX_CONTEXT_WINDOW : LITELLM_STANDARD_CONTEXT_WINDOW;
+		for (const model of this.#models) {
+			if (model.provider === "litellm" && LITELLM_MAX_CONTEXT_MODEL_IDS.has(model.id)) {
+				model.contextWindow = contextWindow;
+			}
+		}
+	}
+
+	/** Update current internal OpenAI context limits without changing subscription models. */
+	setLiteLLMMaxContext(enabled: boolean): void {
+		this.#litellmMaxContext = enabled;
+		this.#applyLiteLLMContextWindow();
+		this.#rebuildCanonicalIndex();
+	}
+
 	/** Load built-in models, applying provider-level overrides only.
 	 *  Per-model overrides are applied later by #applyModelOverrides. */
 	#loadBuiltInModels(overrides: Map<string, ProviderOverride>): Model<Api>[] {
@@ -1433,6 +1456,8 @@ export class ModelRegistry {
 		// Merge runtime extension models so they survive online discovery completion
 		const combined = this.#mergeCustomModels(withConfigModels, this.#runtimeModelOverlays);
 		this.#models = this.#applyProviderModelAllowlists(this.#applyModelOverrides(combined, this.#modelOverrides));
+		this.#applyOpenAICodexContextWindow();
+		this.#applyLiteLLMContextWindow();
 		this.#rebuildCanonicalIndex();
 	}
 

package/src/config/settings-schema.ts

@@ -161,21 +161,22 @@ const DEFAULT_CYCLE_ORDER: string[] = ["smol", "default", "slow"];
 /**
  * Binary-baked default model role. Ships in the binary so a fresh install needs
  * NO `~/.xcsh/agent/config.yml` — `/login` only supplies the (PII) proxy URL + key.
- * GPT-5.6 Sol High is the vision-capable production default through the OpenAI
+ * GPT-5.6 Terra Medium is the current general-purpose default through the OpenAI
  * Chat Completions route on LiteLLM. Keep the effort explicit in the role so returning
  * from a lower-effort role cannot inherit that role's effort.
  */
-export const DEFAULT_MODEL_ROLE = "litellm/gpt-5.6-sol:high";
+export const DEFAULT_MODEL_ROLE = "litellm/gpt-5.6-terra:medium";
 /** Fast role for lightweight work (commit messages, titles, memory summaries). */
-const SMOL_MODEL_ROLE = "litellm/gpt-5.6-sol:low";
+const SMOL_MODEL_ROLE = "litellm/gpt-6-luna:low";
 /**
- * Baked role map. `smol` keeps GPT-5.6 Sol at low effort for latency-sensitive
- * work; both `default` and `slow` restore vision-capable GPT-5.6 Sol High.
+ * Baked role map for the current internal LiteLLM catalog. Sol handles slow and
+ * planning work, while Luna and Terra cover lightweight and default work.
  */
 const DEFAULT_MODEL_ROLES: Record<string, string> = {
 	default: DEFAULT_MODEL_ROLE,
 	smol: SMOL_MODEL_ROLE,
-	slow: DEFAULT_MODEL_ROLE,
+	slow: "litellm/gpt-6-sol:high",
+	plan: "litellm/gpt-6-sol:high",
 };
 const EMPTY_MODEL_TAGS_RECORD: ModelTagsSettings = {};
 export const DEFAULT_BASH_INTERCEPTOR_RULES: BashInterceptorRule[] = [
@@ -2123,6 +2124,16 @@ export const SETTINGS_SCHEMA = {
 		},
 	},
 
+	"providers.litellmMaxContext": {
+		type: "boolean",
+		default: false,
+		ui: {
+			tab: "providers",
+			label: "LiteLLM Maximum Context",
+			description: "Use the verified 1.05M-token context window for current internal OpenAI models",
+		},
+	},
+
 	"providers.parallelFetch": {
 		type: "boolean",
 		default: true,