@f5-sales-demo/xcsh 22.4.11 → 22.5.0

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@f5-sales-demo/xcsh",
-	"version": "22.4.11",
+	"version": "22.5.0",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"homepage": "https://github.com/f5-sales-demo/xcsh",
 	"author": "Can Boluk",
@@ -67,13 +67,13 @@
 	},
 	"dependencies": {
 		"@agentclientprotocol/sdk": "1.4.0",
-		"@f5-sales-demo/pi-agent-core": "22.4.11",
-		"@f5-sales-demo/pi-ai": "22.4.11",
-		"@f5-sales-demo/pi-natives": "22.4.11",
-		"@f5-sales-demo/pi-resource-management": "22.4.11",
-		"@f5-sales-demo/pi-tui": "22.4.11",
-		"@f5-sales-demo/pi-utils": "22.4.11",
-		"@f5-sales-demo/xcsh-stats": "22.4.11",
+		"@f5-sales-demo/pi-agent-core": "22.5.0",
+		"@f5-sales-demo/pi-ai": "22.5.0",
+		"@f5-sales-demo/pi-natives": "22.5.0",
+		"@f5-sales-demo/pi-resource-management": "22.5.0",
+		"@f5-sales-demo/pi-tui": "22.5.0",
+		"@f5-sales-demo/pi-utils": "22.5.0",
+		"@f5-sales-demo/xcsh-stats": "22.5.0",
 		"@mozilla/readability": "^0.6",
 		"@resvg/resvg-wasm": "2.6.2",
 		"@sinclair/typebox": "0.34.52",

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 = 11;
+export const CURRENT_CONFIG_VERSION = 12;
 const LITELLM_CONFIG_DIR_MODE = 0o700;
 const LITELLM_MODELS_FILE_MODE = 0o600;
 const GENERATED_LITELLM_MARKER = "# Auto-generated by xcsh for LiteLLM proxy";
@@ -119,6 +119,7 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"    modelAllowlist:",
 		"      - gpt-6-luna",
 		"      - gpt-5.6-terra",
+		"      - gpt-6.1-sol",
 		"      - gpt-6-sol",
 		"      - gpt-6-astra",
 		"      - gpt-5.6-luna",
@@ -133,6 +134,10 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"        name: GPT-5.6 Terra",
 		"        api: openai-responses",
 		`        baseUrl: "${responsesBaseUrl}"`,
+		"      - id: gpt-6.1-sol",
+		"        name: GPT-6.1 Sol",
+		"        api: openai-responses",
+		`        baseUrl: "${responsesBaseUrl}"`,
 		"      - id: gpt-6-sol",
 		"        name: GPT-6 Sol",
 		"        api: openai-responses",
@@ -198,6 +203,25 @@ export function generateModelsYml(baseUrl: string, options?: GenerateModelsYmlOp
 		"        maxTokens: 128000",
 		"        compat:",
 		"          supportsTemperature: false",
+		"      gpt-6.1-sol:",
+		"        reasoning: true",
+		"        input:",
+		"          - text",
+		"          - image",
+		"        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }",
+		"        thinking:",
+		"          mode: effort",
+		"          defaultLevel: medium",
+		"          supportedLevels:",
+		"            - { 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: 922000",
+		"        maxTokens: 128000",
+		"        compat:",
+		"          supportsTemperature: false",
 		"      gpt-6-sol:",
 		"        reasoning: true",
 		"        input:",

package/src/config/model-catalog.ts

@@ -12,6 +12,7 @@ function compareVersions(left: string, right: string): number {
 
 /** Present the exact model-interaction contract when each provider advertises the model. */
 const CODEX_INTERACTION_MODEL_IDS = new Set([
+	"gpt-6.1-sol",
 	"gpt-6-astra",
 	"gpt-6-sol",
 	"gpt-6-luna",

package/src/config/model-registry.ts

@@ -1,12 +1,15 @@
 import * as path from "node:path";
 import {
+	type AnyModel,
 	type Api,
 	type AssistantMessageEventStream,
 	applyCodexInteractionMetadata,
 	applyGeneratedModelPolicies,
+	type ClassifierModel,
 	type CodexContextTier,
 	type Context,
 	createModelManager,
+	createOperationCatalog,
 	DEFAULT_LOCAL_TOKEN,
 	enrichModelThinking,
 	getBundledModels,
@@ -14,6 +17,7 @@ import {
 	googleAntigravityModelManagerOptions,
 	googleGeminiCliModelManagerOptions,
 	googleVertexModelManagerOptions,
+	type ImageModel,
 	type Model,
 	type ModelManagerOptions,
 	type ModelRefreshStrategy,
@@ -31,7 +35,7 @@ import {
 	unregisterCustomApis,
 	unregisterOAuthProviders,
 } from "@f5-sales-demo/pi-ai";
-import { $env, $envExact, isRecord, logger } from "@f5-sales-demo/pi-utils";
+import { $env, $envExact, getModelDbPath, isRecord, logger } from "@f5-sales-demo/pi-utils";
 import { type Static, Type } from "@sinclair/typebox";
 import { type ConfigError, ConfigFile } from "../config";
 import { hasLiteLLMEnv, probeAndUpgradeLiteLLMConfig, startupHealthCheck } from "../config/auto-config";
@@ -147,6 +151,8 @@ const ReasoningEffortMapSchema = Type.Object({
 	medium: Type.Optional(Type.String()),
 	high: Type.Optional(Type.String()),
 	xhigh: Type.Optional(Type.String()),
+	max: Type.Optional(Type.String()),
+	ultra: Type.Optional(Type.String()),
 });
 
 const OpenAICompatSchema = Type.Object({
@@ -157,6 +163,7 @@ const OpenAICompatSchema = Type.Object({
 	reasoningEffortMap: Type.Optional(ReasoningEffortMapSchema),
 	maxTokensField: Type.Optional(Type.Union([Type.Literal("max_completion_tokens"), Type.Literal("max_tokens")])),
 	supportsUsageInStreaming: Type.Optional(Type.Boolean()),
+	supportsFinishReason: Type.Optional(Type.Boolean()),
 	requiresToolResultName: Type.Optional(Type.Boolean()),
 	requiresAssistantAfterToolResult: Type.Optional(Type.Boolean()),
 	requiresThinkingAsText: Type.Optional(Type.Boolean()),
@@ -167,12 +174,77 @@ const OpenAICompatSchema = Type.Object({
 			Type.Literal("zai"),
 			Type.Literal("qwen"),
 			Type.Literal("qwen-chat-template"),
+			Type.Literal("ant-ling"),
+			Type.Literal("baseten"),
+			Type.Literal("deepseek"),
+			Type.Literal("together"),
+			Type.Literal("chat-template"),
+			Type.Literal("string-thinking"),
 		]),
 	),
 	openRouterRouting: Type.Optional(OpenRouterRoutingSchema),
+	chatTemplateKwargs: Type.Optional(
+		Type.Record(
+			Type.String(),
+			Type.Union([
+				Type.String(),
+				Type.Number(),
+				Type.Boolean(),
+				Type.Null(),
+				Type.Object({
+					$var: Type.Union([
+						Type.Literal("thinking.enabled"),
+						Type.Literal("thinking.effort"),
+						Type.Literal("thinking.budget"),
+					]),
+					omitWhenOff: Type.Optional(Type.Boolean()),
+				}),
+			]),
+		),
+	),
+	chatTemplateArgs: Type.Optional(
+		Type.Record(
+			Type.String(),
+			Type.Union([
+				Type.String(),
+				Type.Number(),
+				Type.Boolean(),
+				Type.Null(),
+				Type.Object({
+					$var: Type.Union([
+						Type.Literal("thinking.enabled"),
+						Type.Literal("thinking.effort"),
+						Type.Literal("thinking.budget"),
+					]),
+					omitWhenOff: Type.Optional(Type.Boolean()),
+				}),
+			]),
+		),
+	),
+	thinkingTokenBudgetField: Type.Optional(
+		Type.Union([
+			Type.Literal("thinking_token_budget"),
+			Type.Literal("thinking_budget"),
+			Type.Literal("thinking_budget_tokens"),
+		]),
+	),
+	supportsThinkingTokenBudget: Type.Optional(Type.Boolean()),
+	sendSessionAffinityHeaders: Type.Optional(Type.Boolean()),
+	allowEmptySignature: Type.Optional(Type.Boolean()),
+	supportsCacheControlOnTools: Type.Optional(Type.Boolean()),
+	forceAdaptiveThinking: Type.Optional(Type.Boolean()),
 	vercelGatewayRouting: Type.Optional(VercelGatewayRoutingSchema),
 	extraBody: Type.Optional(Type.Record(Type.String(), Type.Unknown())),
 	supportsStrictMode: Type.Optional(Type.Boolean()),
+	supportsMidConvoSystemMessages: Type.Optional(Type.Boolean()),
+	supportsMidConvoToolChanges: Type.Optional(Type.Boolean()),
+	supportsOpenAIGrammarTools: Type.Optional(Type.Boolean()),
+	supportsAdditionalTools: Type.Optional(Type.Boolean()),
+	supportsToolSearch: Type.Optional(Type.Boolean()),
+	supportsExplicitPromptCacheMode: Type.Optional(Type.Boolean()),
+	supportsCachedReasoningUpdates: Type.Optional(Type.Boolean()),
+	supportsWebSocketSteering: Type.Optional(Type.Boolean()),
+	supportsMaxOutputTokens: Type.Optional(Type.Boolean()),
 	toolStrictMode: Type.Optional(Type.Union([Type.Literal("all_strict"), Type.Literal("none")])),
 });
 
@@ -184,6 +256,7 @@ const EffortSchema = Type.Union([
 	Type.Literal("high"),
 	Type.Literal("xhigh"),
 	Type.Literal("max"),
+	Type.Literal("ultra"),
 ]);
 
 const ThinkingControlModeSchema = Type.Union([
@@ -213,6 +286,8 @@ const ModelDefinitionSchema = Type.Object({
 	name: Type.Optional(Type.String({ minLength: 1 })),
 	api: Type.Optional(
 		Type.Union([
+			Type.Literal("pi-messages"),
+			Type.Literal("mistral-conversations"),
 			Type.Literal("openai-completions"),
 			Type.Literal("openai-responses"),
 			Type.Literal("openai-codex-responses"),
@@ -232,10 +307,22 @@ const ModelDefinitionSchema = Type.Object({
 			output: Type.Number(),
 			cacheRead: Type.Number(),
 			cacheWrite: Type.Number(),
+			tiers: Type.Optional(
+				Type.Array(
+					Type.Object({
+						inputTokensAbove: Type.Number(),
+						input: Type.Number(),
+						output: Type.Number(),
+						cacheRead: Type.Number(),
+						cacheWrite: Type.Number(),
+					}),
+				),
+			),
 		}),
 	),
 	premiumMultiplier: Type.Optional(Type.Number()),
 	contextWindow: Type.Optional(Type.Number()),
+	maxInputTokens: Type.Optional(Type.Number()),
 	maxTokens: Type.Optional(Type.Number()),
 	headers: Type.Optional(Type.Record(Type.String(), Type.String())),
 	compat: Type.Optional(OpenAICompatSchema),
@@ -254,10 +341,22 @@ const ModelOverrideSchema = Type.Object({
 			output: Type.Optional(Type.Number()),
 			cacheRead: Type.Optional(Type.Number()),
 			cacheWrite: Type.Optional(Type.Number()),
+			tiers: Type.Optional(
+				Type.Array(
+					Type.Object({
+						inputTokensAbove: Type.Number(),
+						input: Type.Number(),
+						output: Type.Number(),
+						cacheRead: Type.Number(),
+						cacheWrite: Type.Number(),
+					}),
+				),
+			),
 		}),
 	),
 	premiumMultiplier: Type.Optional(Type.Number()),
 	contextWindow: Type.Optional(Type.Number()),
+	maxInputTokens: Type.Optional(Type.Number()),
 	maxTokens: Type.Optional(Type.Number()),
 	headers: Type.Optional(Type.Record(Type.String(), Type.String())),
 	compat: Type.Optional(OpenAICompatSchema),
@@ -329,6 +428,7 @@ interface ProviderValidationModel {
 	id: string;
 	api?: Api;
 	contextWindow?: number;
+	maxInputTokens?: number;
 	maxTokens?: number;
 }
 
@@ -543,10 +643,12 @@ type OllamaDiscoveredModelMetadata = {
 	reasoning: boolean;
 	input: ("text" | "image")[];
 	contextWindow?: number;
+	maxInputTokens?: number;
 };
 
 type LlamaCppDiscoveredServerMetadata = {
 	contextWindow?: number;
+	maxInputTokens?: number;
 	input?: ("text" | "image")[];
 };
 
@@ -699,6 +801,7 @@ function applyModelOverride(model: Model<Api>, override: ModelOverride): Model<A
 	if (override.thinking !== undefined) result.thinking = override.thinking as ThinkingConfig;
 	if (override.input !== undefined) result.input = override.input as ("text" | "image")[];
 	if (override.contextWindow !== undefined) result.contextWindow = override.contextWindow;
+	if (override.maxInputTokens !== undefined) result.maxInputTokens = override.maxInputTokens;
 	if (override.maxTokens !== undefined) result.maxTokens = override.maxTokens;
 	if (override.contextPromotionTarget !== undefined) result.contextPromotionTarget = override.contextPromotionTarget;
 	if (override.premiumMultiplier !== undefined) result.premiumMultiplier = override.premiumMultiplier;
@@ -708,6 +811,7 @@ function applyModelOverride(model: Model<Api>, override: ModelOverride): Model<A
 			output: override.cost.output ?? model.cost.output,
 			cacheRead: override.cost.cacheRead ?? model.cost.cacheRead,
 			cacheWrite: override.cost.cacheWrite ?? model.cost.cacheWrite,
+			tiers: override.cost.tiers ?? model.cost.tiers,
 		};
 	}
 	if (override.headers) {
@@ -725,8 +829,9 @@ interface CustomModelDefinitionLike {
 	reasoning?: boolean;
 	thinking?: ThinkingConfig;
 	input?: ("text" | "image")[];
-	cost?: { input: number; output: number; cacheRead: number; cacheWrite: number };
+	cost?: import("@f5-sales-demo/pi-ai").ModelCost;
 	contextWindow?: number;
+	maxInputTokens?: number;
 	maxTokens?: number;
 	headers?: Record<string, string>;
 	compat?: Model<Api>["compat"];
@@ -747,8 +852,9 @@ type CustomModelOverlay = {
 	reasoning?: boolean;
 	thinking?: ThinkingConfig;
 	input?: ("text" | "image")[];
-	cost?: { input: number; output: number; cacheRead: number; cacheWrite: number };
+	cost?: import("@f5-sales-demo/pi-ai").ModelCost;
 	contextWindow?: number;
+	maxInputTokens?: number;
 	maxTokens?: number;
 	headers?: Record<string, string>;
 	compat?: Model<Api>["compat"];
@@ -795,6 +901,7 @@ function buildCustomModelOverlay(
 		input: modelDef.input as ("text" | "image")[] | undefined,
 		cost: modelDef.cost,
 		contextWindow: modelDef.contextWindow,
+		maxInputTokens: modelDef.maxInputTokens,
 		maxTokens: modelDef.maxTokens,
 		headers: mergeCustomModelHeaders(providerHeaders, modelDef.headers, authHeader, providerApiKey),
 		compat: mergeCompat(providerCompat, modelDef.compat),
@@ -825,6 +932,7 @@ function finalizeCustomModel(model: CustomModelOverlay, options: CustomModelBuil
 		thinking: resolvedModel.thinking,
 		input: input as ("text" | "image")[],
 		cost,
+		maxInputTokens: resolvedModel.maxInputTokens,
 		contextWindow: resolvedModel.contextWindow ?? (options.useDefaults ? 128000 : undefined),
 		maxTokens: resolvedModel.maxTokens ?? (options.useDefaults ? 16384 : undefined),
 		headers: resolvedModel.headers,
@@ -854,6 +962,7 @@ function getDisabledProviderIdsFromSettings(): Set<string> {
  * Model registry - loads and manages models, resolves API keys via AuthStorage.
  */
 export class ModelRegistry {
+	#operationCatalog: ReturnType<typeof createOperationCatalog>;
 	#models: Model<Api>[] = [];
 	#canonicalIndex: CanonicalModelIndex = { records: [], byId: new Map(), bySelector: new Map() };
 	#customProviderApiKeys: Map<string, string> = new Map();
@@ -907,6 +1016,9 @@ export class ModelRegistry {
 	) {
 		this.#modelsConfigFile = ModelsConfigFile.relocate(modelsPath);
 		this.#cacheDbPath = modelsPath ? path.join(path.dirname(modelsPath), "models.db") : undefined;
+		this.#operationCatalog = createOperationCatalog({
+			cacheDbPath: this.#cacheDbPath ?? getModelDbPath(),
+		});
 		// Set up fallback resolver for custom provider API keys
 		this.authStorage.setFallbackResolver(provider => {
 			const keyConfig = this.#customProviderApiKeys.get(provider);
@@ -2066,6 +2178,7 @@ export class ModelRegistry {
 		const items: Array<{
 			id: string;
 			contextWindow?: number;
+			maxInputTokens?: number;
 			context_window?: unknown;
 			max_input_tokens?: unknown;
 			max_output_tokens?: unknown;
@@ -2233,6 +2346,42 @@ export class ModelRegistry {
 		return this.#models;
 	}
 
+	getImageModels(provider: string): ImageModel[] {
+		return this.#operationCatalog
+			.getImageModels(provider)
+			.map(model => ({ ...model, baseUrl: this.#providerOverrides.get(provider)?.baseUrl ?? model.baseUrl }));
+	}
+
+	getClassifierModels(provider: string): ClassifierModel[] {
+		return this.#operationCatalog
+			.getClassifierModels(provider)
+			.map(model => ({ ...model, baseUrl: this.#providerOverrides.get(provider)?.baseUrl ?? model.baseUrl }));
+	}
+
+	getAnyModels(provider: string): AnyModel[] {
+		return [
+			...this.#models.filter(model => model.provider === provider),
+			...this.getImageModels(provider),
+			...this.getClassifierModels(provider),
+		];
+	}
+
+	async refreshOperationModels(provider: string, signal?: AbortSignal): Promise<void> {
+		const apiKey = await this.getApiKeyForProvider(provider);
+		await this.#operationCatalog.refresh(provider, {
+			apiKey,
+			signal,
+			baseUrl: this.getProviderBaseUrl(provider),
+			...(provider === "llama.cpp"
+				? {
+						modelIds: this.getAll()
+							.filter(model => model.provider === provider)
+							.map(model => model.id),
+					}
+				: {}),
+		});
+	}
+
 	#isModelAvailable(model: Model<Api>): boolean {
 		const disabledProviders = getDisabledProviderIdsFromSettings();
 		return (
@@ -2711,7 +2860,7 @@ export interface ProviderConfigInput {
 		reasoning: boolean;
 		thinking?: ThinkingConfig;
 		input: ("text" | "image")[];
-		cost: { input: number; output: number; cacheRead: number; cacheWrite: number };
+		cost: import("@f5-sales-demo/pi-ai").ModelCost;
 		contextWindow: number;
 		maxTokens: number;
 		headers?: Record<string, string>;

package/src/config/settings-schema.ts

@@ -161,22 +161,21 @@ 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 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
+ * GPT-6.1 Sol Medium is the default through the OpenAI Responses 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-terra:medium";
+export const DEFAULT_MODEL_ROLE = "litellm/gpt-6.1-sol:medium";
 /** Fast role for lightweight work (commit messages, titles, memory summaries). */
 const SMOL_MODEL_ROLE = "litellm/gpt-6-luna:low";
 /**
  * Baked role map for the current internal LiteLLM catalog. Sol handles slow and
- * planning work, while Luna and Terra cover lightweight and default work.
+ * planning and default work, while Luna covers lightweight work.
  */
 export const DEFAULT_MODEL_ROLES: Record<string, string> = {
 	default: DEFAULT_MODEL_ROLE,
 	smol: SMOL_MODEL_ROLE,
-	slow: "litellm/gpt-6-sol:high",
-	plan: "litellm/gpt-6-sol:high",
+	slow: "litellm/gpt-6.1-sol:high",
+	plan: "litellm/gpt-6.1-sol:high",
 };
 const EMPTY_MODEL_TAGS_RECORD: ModelTagsSettings = {};
 export const DEFAULT_BASH_INTERCEPTOR_RULES: BashInterceptorRule[] = [

package/src/extensibility/custom-tools/types.ts

@@ -184,6 +184,9 @@ export interface CustomTool<TParams extends TSchema = TSchema, TDetails = any> {
 	hidden?: boolean;
 	/** If true, tool may stage deferred changes that require explicit resolve/discard. */
 	deferrable?: boolean;
+	/** Opt-in execution through the host's background job lifecycle. */
+	async?: boolean;
+	constrainedSampling?: import("@f5-sales-demo/pi-ai").Tool["constrainedSampling"];
 	/**
 	 * Execute the tool.
 	 * @param toolCallId - Unique ID for this tool call

package/src/extensibility/extensions/types.ts

@@ -313,6 +313,8 @@ export interface ToolSessionEvent {
  * Tool definition for registerTool().
  */
 export interface ToolDefinition<TParams extends TSchema = TSchema, TDetails = unknown> {
+	async?: boolean;
+	constrainedSampling?: import("@f5-sales-demo/pi-ai").Tool["constrainedSampling"];
 	/** Tool name (used in LLM tool calls) */
 	name: string;
 	/** Human-readable label for UI */
@@ -1323,7 +1325,7 @@ export interface ProviderModelConfig {
 	/** Supported input types. */
 	input: ("text" | "image")[];
 	/** Cost per million tokens. */
-	cost: { input: number; output: number; cacheRead: number; cacheWrite: number };
+	cost: import("@f5-sales-demo/pi-ai").ModelCost;
 	/** Premium Copilot requests charged per user-initiated request. */
 	premiumMultiplier?: number;
 	/** Maximum context window size in tokens. */