@oh-my-pi/pi-coding-agent 16.2.11 → 16.2.12

package/src/cli/bench-cli.ts

@@ -14,12 +14,12 @@ import type {
 	SimpleStreamOptions,
 } from "@oh-my-pi/pi-ai";
 import { resolveModelServiceTier, streamSimple } from "@oh-my-pi/pi-ai";
-import { buildModelProviderPriorityRank, type CanonicalModelVariant } from "@oh-my-pi/pi-catalog/identity";
+import { buildModelProviderPriorityRank } from "@oh-my-pi/pi-catalog/identity";
 import { replaceTabs, truncateToWidth } from "@oh-my-pi/pi-tui";
 import { formatDuration, getProjectDir } from "@oh-my-pi/pi-utils";
 import chalk from "chalk";
 import type { ApiKeyResolverModel } from "../config/api-key-resolver";
-import { type CanonicalModelQueryOptions, ModelRegistry } from "../config/model-registry";
+import { ModelRegistry } from "../config/model-registry";
 import {
 	formatModelSelectorValue,
 	formatModelString,
@@ -55,9 +55,6 @@ export interface BenchModelRegistry {
 	getAll(): Model<Api>[];
 	getApiKey(model: Model<Api>, sessionId?: string): Promise<string | undefined>;
 	resolver(model: ApiKeyResolverModel, sessionId?: string): ApiKeyResolver;
-	resolveCanonicalModel?(canonicalId: string, options?: CanonicalModelQueryOptions): Model<Api> | undefined;
-	getCanonicalVariants?(canonicalId: string, options?: CanonicalModelQueryOptions): CanonicalModelVariant[];
-	getCanonicalId?(model: Model<Api>): string | undefined;
 	hasConfiguredAuth?(model: Model<Api>): boolean;
 }
 
@@ -440,13 +437,7 @@ function resolveAuthenticatedAlternative(
 		seen.add(key);
 		if (modelRegistry.hasConfiguredAuth?.(candidate)) authenticated.push(candidate);
 	};
-	// Canonical variants link the same logical model across providers even when
-	// ids differ (e.g. fireworks `gpt-oss-20b` <-> openrouter `openai/gpt-oss-20b`).
-	const canonicalId = modelRegistry.getCanonicalId?.(model);
-	if (canonicalId) {
-		for (const variant of modelRegistry.getCanonicalVariants?.(canonicalId) ?? []) consider(variant.model);
-	}
-	// Same-id fallback for entries outside the canonical index.
+	// Same-id fallback for equivalent entries under providers with configured auth.
 	for (const candidate of modelRegistry.getAll()) {
 		if (candidate.id === model.id) consider(candidate);
 	}

package/src/cli/dry-balance-cli.ts

@@ -12,11 +12,10 @@ import type {
 	SimpleStreamOptions,
 } from "@oh-my-pi/pi-ai";
 import { streamSimple } from "@oh-my-pi/pi-ai";
-import type { CanonicalModelVariant } from "@oh-my-pi/pi-catalog/identity";
 import { replaceTabs, truncateToWidth } from "@oh-my-pi/pi-tui";
 import { formatDuration, getProjectDir } from "@oh-my-pi/pi-utils";
 import chalk from "chalk";
-import { type CanonicalModelQueryOptions, ModelRegistry } from "../config/model-registry";
+import { ModelRegistry } from "../config/model-registry";
 import {
 	formatModelString,
 	getModelMatchPreferences,
@@ -74,9 +73,6 @@ export interface DryBalanceModelRegistry {
 	getAll(): Model<Api>[];
 	getAvailable(): Model<Api>[];
 	getApiKey(model: Model<Api>, sessionId?: string): Promise<string | undefined>;
-	getCanonicalVariants(canonicalId: string, options?: CanonicalModelQueryOptions): CanonicalModelVariant[];
-	resolveCanonicalModel?(canonicalId: string, options?: CanonicalModelQueryOptions): Model<Api> | undefined;
-	getCanonicalId?(model: Model<Api>): string | undefined;
 }
 
 export interface DryBalanceRuntime {
@@ -566,7 +562,6 @@ async function resolveDryBalanceModel(
 	const defaultRoleSpec = resolveModelRoleValue(settings?.getModelRole("default"), allowedModels, {
 		settings,
 		matchPreferences: preferences,
-		modelRegistry,
 	});
 	if (defaultRoleSpec.model) {
 		return { model: defaultRoleSpec.model, warning: defaultRoleSpec.warning };

package/src/cli/models-cli.ts

@@ -2,7 +2,7 @@
  * `omp models` — list, search, and refresh available models.
  *
  * Subcommands:
- * - `ls` (default): list every available model with a canonical + provider view.
+ * - `ls` (default): list every available model grouped by provider.
  * - `find <substring>`: list models whose provider, id, or name contains the substring.
  * - `refresh`: force an online catalog re-fetch (ignoring the model cache TTL),
  *   then list. This is the supported replacement for `rm -rf ~/.omp/models.db`
@@ -21,7 +21,7 @@ import { discoverAndLoadExtensions, loadExtensions } from "../extensibility/exte
 import { discoverAuthStorage } from "../sdk";
 import { EventBus } from "../utils/event-bus";
 
-export type ModelsAction = "ls" | "find" | "refresh" | "canonical";
+export type ModelsAction = "ls" | "find" | "refresh";
 
 export interface ModelsCommandArgs {
 	action: ModelsAction;
@@ -48,7 +48,6 @@ const KNOWN_ACTIONS: Record<string, ModelsAction> = {
 	list: "ls",
 	find: "find",
 	refresh: "refresh",
-	canonical: "canonical",
 };
 
 /** Resolve the two positional args into an action + filter (provider names fall through to `ls`). */
@@ -77,22 +76,10 @@ interface ModelJson {
 	cost: Model<Api>["cost"];
 }
 
-interface CanonicalJson {
-	id: string;
-	selected: string;
-	variants: number;
-	contextWindow: number | null;
-	maxTokens: number | null;
-}
-
 interface ModelsJson {
 	models: ModelJson[];
 }
 
-interface CanonicalModelsJson {
-	canonical: CanonicalJson[];
-}
-
 function writeLine(line = ""): void {
 	process.stdout.write(`${line}\n`);
 }
@@ -253,67 +240,6 @@ function renderProviderModels(
 	}
 }
 
-/** `omp models canonical`: the coalesced canonical view (one row per canonical id). */
-function renderCanonicalModels(modelRegistry: ModelRegistry, pattern: string | undefined, json: boolean): void {
-	const selections = modelRegistry.getCanonicalModelSelections({ availableOnly: true });
-	const needle = pattern?.toLowerCase();
-	const filtered = needle
-		? selections.filter(
-				({ record, model }) =>
-					record.id.toLowerCase().includes(needle) ||
-					`${model.provider}/${model.id}`.toLowerCase().includes(needle),
-			)
-		: selections;
-
-	if (json) {
-		const output: CanonicalModelsJson = {
-			canonical: filtered
-				.map(({ record, model }) => ({
-					id: record.id,
-					selected: `${model.provider}/${model.id}`,
-					variants: record.variants.length,
-					contextWindow: model.contextWindow,
-					maxTokens: model.maxTokens,
-				}))
-				.sort((left, right) => left.id.localeCompare(right.id)),
-		};
-		writeLine(JSON.stringify(output));
-		return;
-	}
-
-	if (selections.length === 0) {
-		writeLine("No models available. Set API keys in environment variables.");
-		return;
-	}
-	if (filtered.length === 0) {
-		writeLine(`No canonical models matching "${pattern}"`);
-		return;
-	}
-
-	const rows = filtered
-		.slice()
-		.sort((left, right) => left.record.id.localeCompare(right.record.id))
-		.map(({ record, model }) => [
-			record.id,
-			`${model.provider}/${model.id}`,
-			String(record.variants.length),
-			formatLimit(model.contextWindow),
-			formatLimit(model.maxTokens),
-		]);
-	for (const line of boxTable(
-		[
-			{ header: "canonical" },
-			{ header: "selected" },
-			{ header: "variants", align: "right" },
-			{ header: "context", align: "right" },
-			{ header: "max-out", align: "right" },
-		],
-		rows,
-	)) {
-		writeLine(line);
-	}
-}
-
 /**
  * Options for {@link runModelsListing}: render the catalog from a caller-supplied
  * registry. Loads extensions (CLI `-e` paths and configured `settings.extensions`)
@@ -376,11 +302,7 @@ export async function runModelsListing(options: RunModelsListingOptions): Promis
 	// Discover runtime (extension) provider catalogs now that they are registered.
 	await modelRegistry.refreshRuntimeProviders(action === "refresh" ? "online" : "online-if-uncached");
 
-	if (action === "canonical") {
-		renderCanonicalModels(modelRegistry, pattern, json);
-	} else {
-		renderProviderModels(modelRegistry, action, pattern, json);
-	}
+	renderProviderModels(modelRegistry, action, pattern, json);
 }
 
 /**

package/src/commands/models.ts

@@ -10,7 +10,7 @@ export default class Models extends Command {
 
 	static args = {
 		action: Args.string({
-			description: "ls (default) | find | refresh | canonical | <provider>",
+			description: "ls (default) | find | refresh | <provider>",
 			required: false,
 		}),
 		pattern: Args.string({
@@ -40,7 +40,6 @@ export default class Models extends Command {
 		`# List one provider's models (any provider name works)\n  ${APP_NAME} models openai-codex`,
 		`# Find models by substring\n  ${APP_NAME} models find minimax`,
 		`# Force a fresh catalog fetch (replaces rm -rf ~/.omp/models.db)\n  ${APP_NAME} models refresh`,
-		`# Show the coalesced canonical model view\n  ${APP_NAME} models canonical`,
 		`# Machine-readable output\n  ${APP_NAME} models --json`,
 	];
 

package/src/commit/model-selection.ts

@@ -36,8 +36,8 @@ export async function resolvePrimaryModel(
 	const available = modelRegistry.getAvailable();
 	const matchPreferences = getModelMatchPreferences(settings);
 	const resolved = override
-		? resolveModelRoleValue(override, available, { settings, matchPreferences, modelRegistry })
-		: resolveRoleSelection(["commit", "smol", ...MODEL_ROLE_IDS], settings, available, modelRegistry);
+		? resolveModelRoleValue(override, available, { settings, matchPreferences })
+		: resolveRoleSelection(["commit", "smol", ...MODEL_ROLE_IDS], settings, available);
 	const model = resolved?.model;
 	if (!model) {
 		throw new Error("No model available for commit generation");
@@ -60,7 +60,7 @@ export async function resolveSmolModel(
 	fallbackApiKey: ApiKey,
 ): Promise<ResolvedCommitModel> {
 	const available = modelRegistry.getAvailable();
-	const resolvedSmol = resolveRoleSelection(["smol"], settings, available, modelRegistry);
+	const resolvedSmol = resolveRoleSelection(["smol"], settings, available);
 	if (resolvedSmol?.model) {
 		const apiKey = await modelRegistry.getApiKey(resolvedSmol.model);
 		if (apiKey) {
@@ -74,7 +74,7 @@ export async function resolveSmolModel(
 
 	const matchPreferences = getModelMatchPreferences(settings);
 	for (const pattern of MODEL_PRIO.smol) {
-		const candidate = parseModelPattern(pattern, available, matchPreferences, { modelRegistry }).model;
+		const candidate = parseModelPattern(pattern, available, matchPreferences).model;
 		if (!candidate) continue;
 		const apiKey = await modelRegistry.getApiKey(candidate);
 		if (apiKey) {

package/src/config/inline-tool-descriptors-mode.ts

@@ -9,8 +9,7 @@ import { modelFamilyToken } from "@oh-my-pi/pi-catalog/identity";
  * Gemini benefits from descriptors in-prompt; other providers keep them in the
  * tool schemas. `on`/`off` are explicit user overrides.
  *
- * @param modelId Canonical model id (e.g. `gemini-3-pro`); resolve aliases via
- *   `ModelRegistry.getCanonicalId` before calling so `auto` classifies correctly.
+ * @param modelId Model id (e.g. `gemini-3-pro`) used to classify `auto`.
  */
 export function shouldInlineToolDescriptors(
 	setting: "auto" | "on" | "off" | undefined,

package/src/config/model-discovery.ts

@@ -19,7 +19,7 @@ import {
 	OPENAI_COMPAT_DISCOVERY_DEFAULT_CONTEXT_WINDOW,
 	OPENAI_COMPAT_DISCOVERY_DEFAULT_MAX_TOKENS,
 } from "@oh-my-pi/pi-catalog/provider-models/openai-compat";
-import type { ModelSpec } from "@oh-my-pi/pi-catalog/types";
+import type { ModelSpec, OpenAICompat } from "@oh-my-pi/pi-catalog/types";
 import { isRecord } from "@oh-my-pi/pi-utils";
 import type { ProviderDiscovery } from "./models-config-schema";
 
@@ -585,34 +585,60 @@ export async function discoverOpenAIModelsList(
 		? await withAuth(apiKey, key => attempt({ ...baseHeaders, Authorization: `Bearer ${key}` }))
 		: await attempt(baseHeaders);
 	const models = payload.data ?? [];
+	const references = getBundledModelReferenceIndex();
 	const discovered: Model<Api>[] = [];
 	for (const item of models) {
 		const id = item.id;
 		if (!id) continue;
 		const nativeMetadataForModel = nativeMetadata?.get(id);
+		// Thin OpenAI-compatible proxies frequently omit `context_length`/
+		// `max_model_len` on `/v1/models`, leaving discovered models pinned at
+		// the 128K default even when the underlying model is e.g. a proxied
+		// Claude with a 1M window. Resolve the id against the bundled catalog
+		// (same pattern as `discoverProxyModels` and `discoverLiteLLMModels`) so
+		// intrinsic metadata — context/output limits, display name, modality,
+		// reasoning support — flows through when the provider is silent. Local
+		// runtime state and provider-reported values still win; proxy-specific
+		// headers/baseUrl/cost stay local.
+		const reference = resolveModelReference(id, references) as ModelSpec<Api> | undefined;
+		const referenceCompat = reference?.compat as OpenAICompat | undefined;
 		const contextWindow =
 			toPositiveNumberOrUndefined(item.max_model_len) ??
 			toPositiveNumberOrUndefined(item.context_length) ??
 			nativeMetadataForModel?.contextWindow ??
+			reference?.contextWindow ??
 			DISCOVERY_DEFAULT_CONTEXT_WINDOW;
 		discovered.push(
 			buildModel({
 				id,
-				name: id,
+				name: reference?.name ?? id,
 				api: providerConfig.api,
 				provider: providerConfig.provider,
 				baseUrl,
-				reasoning: false,
-				input: nativeMetadataForModel?.input ?? ["text"],
+				reasoning: reference?.reasoning ?? false,
+				thinking: reference?.thinking,
+				input: nativeMetadataForModel?.input ?? reference?.input ?? ["text"],
 				...(providerConfig.discovery.type === "lm-studio" ? { imageInputDecoder: "stb" as const } : {}),
+				// Proxy/gateway pricing is provider-specific and rarely matches
+				// upstream bundled catalogs, so keep costs local-unknown even
+				// when we successfully recover the upstream model identity.
 				cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
 				contextWindow,
-				maxTokens: Math.min(contextWindow, discoveryDefaultMaxTokens(providerConfig.api)),
+				// Cap the reference's output limit at the discovered context
+				// window so an ID collision with a larger bundled model can
+				// never request more tokens than the local runtime advertises.
+				maxTokens: Math.min(reference?.maxTokens ?? discoveryDefaultMaxTokens(providerConfig.api), contextWindow),
 				headers,
 				compat: {
 					supportsStore: false,
 					supportsDeveloperRole: false,
-					supportsReasoningEffort: false,
+					supportsReasoningEffort: referenceCompat?.supportsReasoningEffort ?? false,
+					...(referenceCompat?.reasoningEffortMap
+						? { reasoningEffortMap: referenceCompat.reasoningEffortMap }
+						: {}),
+					...(referenceCompat?.omitReasoningEffort !== undefined
+						? { omitReasoningEffort: referenceCompat.omitReasoningEffort }
+						: {}),
 				},
 			} as ModelSpec<Api>),
 		);