@oh-my-pi/pi-catalog 17.1.4 → 17.1.6

package/CHANGELOG.md

@@ -2,6 +2,19 @@
 
 ## [Unreleased]
 
+## [17.1.6] - 2026-07-27
+
+### Added
+
+- Added SiliconFlow providers (`siliconflow`, `siliconflow-cn`) with dynamic-only OpenAI-compatible model discovery: no bundled catalog — the model list is fetched live from each region's `/v1/models` endpoint, with non-chat entries (embedding, reranker, image, audio, video) filtered out. Discovery hydrates pricing, context/output limits, and reasoning metadata from the provider's models.dev catalog at runtime (with bundled upstream references as a reasoning-only fallback for ids models.dev has not indexed), so reasoning models keep thinking enabled and sessions compact against real context windows. `SILICONFLOW_API_KEY` / `SILICONFLOW_CN_API_KEY` environment variables are wired into `getEnvApiKey`.
+
+## [17.1.5] - 2026-07-27
+
+### Fixed
+
+- Fixed Kimi Code (`kimi-code`) reporting `maxTokens: 32000` for every model — its `/coding/v1/models` discovery mapper and the bundled catalog applied a blanket constant, truncating `k3`/`k3-256k` output at ~4x below their real 131072 ceiling and `kimi-for-coding`/`kimi-for-coding-highspeed` below their 32768 ceiling. Output caps are now derived per family, and the model cache is invalidated so upgrades drop the stale `maxTokens: 32000` rows (including the discovery-only `k3-256k`) instead of serving them until the next network refresh ([#6711](https://github.com/can1357/oh-my-pi/issues/6711)).
+- Fixed Anthropic model discovery 404ing when the registry derived the provider base URL from a bundled model without the `/v1` suffix (`https://api.anthropic.com/models` instead of `/v1/models`), which let a stale text-only cache row shadow fresh models.dev vision metadata — surfacing as snapcompact refusing to run on `claude-opus-5`. Discovery now always targets `/v1/models` while model rows keep the provider base URL ([#6563](https://github.com/can1357/oh-my-pi/issues/6563)).
+
 ## [17.1.4] - 2026-07-26
 
 ### Added

package/dist/types/compat/anthropic.d.ts

@@ -7,5 +7,12 @@ import type { ModelSpec, ResolvedAnthropicCompat } from "../types.js";
  * prefix check would accept lookalikes like `https://api.anthropic.com.evil.com`.
  */
 export declare function isOfficialAnthropicApiUrl(baseUrl?: string): boolean;
+/**
+ * Known non-official URLs that enforce Anthropic thinking signatures on replay.
+ *
+ * Runtime routing calls this with the effective URL because a model's resolved
+ * compat can be stale after Foundry or a provider base-URL override reroutes it.
+ */
+export declare function isAnthropicSigningProxyUrl(baseUrl?: string): boolean;
 /** Build the resolved anthropic-messages compat record for a model spec. */
 export declare function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): ResolvedAnthropicCompat;

package/dist/types/provider-models/bundled-references.d.ts

@@ -8,4 +8,6 @@ import type { Api, Model, ModelSpec } from "../types.js";
  */
 export declare function toModelSpec<TApi extends Api>(model: Model<TApi>): ModelSpec<TApi>;
 export declare function createBundledReferenceMap<TApi extends Api>(provider: Parameters<typeof getBundledModels>[0]): Map<string, ModelSpec<TApi>>;
-export declare function createReferenceResolver<TApi extends Api>(providerRefs: Map<string, ModelSpec<TApi>>): (modelId: string) => ModelSpec<TApi> | undefined;
+type ProviderReferenceSource<TApi extends Api> = Map<string, ModelSpec<TApi>> | (() => Map<string, ModelSpec<TApi>>);
+export declare function createReferenceResolver<TApi extends Api>(providerReferenceSource: ProviderReferenceSource<TApi>): (modelId: string) => ModelSpec<TApi> | undefined;
+export {};

package/dist/types/provider-models/cache-provider-id.d.ts

@@ -0,0 +1,7 @@
+export interface ModelCacheProviderIdOptions {
+    apiKey?: string;
+    baseUrl?: string;
+}
+export declare function getDefaultModelDiscoveryBaseUrl(providerId: string): string | undefined;
+/** Resolve the cache namespace used by a provider's model-manager options without constructing those options. */
+export declare function resolveModelCacheProviderId(providerId: string, options?: ModelCacheProviderIdOptions): string;

package/dist/types/provider-models/descriptors.d.ts

@@ -323,6 +323,18 @@ export declare const CATALOG_PROVIDERS: readonly [{
     readonly catalogDiscovery: {
         readonly label: "Sakana AI";
     };
+}, {
+    readonly id: "siliconflow";
+    readonly defaultModel: "zai-org/GLM-5.1";
+    readonly envVars: readonly ["SILICONFLOW_API_KEY"];
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
+}, {
+    readonly id: "siliconflow-cn";
+    readonly defaultModel: "deepseek-ai/DeepSeek-V4-Pro";
+    readonly envVars: readonly ["SILICONFLOW_CN_API_KEY"];
+    readonly createModelManagerOptions: (config: ModelManagerConfig) => import("../index.js").ModelManagerOptions<"openai-completions", unknown>;
+    readonly dynamicModelsAuthoritative: true;
 }, {
     readonly id: "synthetic";
     readonly defaultModel: "hf:zai-org/GLM-5.1";

package/dist/types/provider-models/index.d.ts

@@ -1,3 +1,4 @@
+export * from "./cache-provider-id.js";
 export * from "./descriptor-types.js";
 export * from "./descriptors.js";
 export * from "./google.js";

package/dist/types/provider-models/openai-compat.d.ts

@@ -166,6 +166,14 @@ export interface DeepSeekModelManagerConfig {
     fetch?: FetchImpl;
 }
 export declare function deepseekModelManagerOptions(config?: DeepSeekModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export interface SiliconFlowModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+export declare function isLikelySiliconFlowChatModelId(id: string): boolean;
+export declare function siliconflowModelManagerOptions(config?: SiliconFlowModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export declare function siliconflowCnModelManagerOptions(config?: SiliconFlowModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface ZhipuCodingPlanModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;
@@ -319,6 +327,22 @@ export interface KimiCodeModelManagerConfig {
     baseUrl?: string;
     fetch?: FetchImpl;
 }
+/**
+ * Kimi Code output ceilings by model family. The `/coding/v1/models` discovery
+ * envelope carries no output-limit field, so the mapper supplies the documented
+ * per-family caps instead of a blanket constant. Values match models.dev's
+ * `kimi-for-coding` and `moonshotai` kimi-k3 entries. See #6711.
+ */
+export declare const KIMI_CODE_K3_MAX_TOKENS = 131072;
+export declare const KIMI_CODE_FOR_CODING_MAX_TOKENS = 32768;
+/** Fallback output cap for Kimi Code families without a documented ceiling (legacy K2 discovery rows). */
+export declare const KIMI_CODE_DEFAULT_MAX_TOKENS = 32000;
+/**
+ * Resolve a Kimi Code model's output ceiling from its id: `k3` / `k3-256k` ->
+ * 131072, `kimi-for-coding[-highspeed]` -> 32768, everything else -> `fallback`.
+ */
+export declare function kimiCodeMaxTokens(modelId: string, fallback?: number): number;
+export declare function kimiCodeMaxTokens(modelId: string, fallback: number | null): number | null;
 export declare function kimiCodeModelManagerOptions(config?: KimiCodeModelManagerConfig): ModelManagerOptions<"openai-completions">;
 /** Native LM Studio metadata keyed by model id from `/api/v0/models`. */
 export interface LmStudioNativeModelMetadata {

package/dist/types/types.d.ts

@@ -432,6 +432,20 @@ export interface AnthropicCompat {
      * blocks instead of normal `tool_use` calls.
      */
     escapeBuiltinToolNames?: boolean;
+    /**
+     * The configured endpoint enforces Anthropic's signature protocol on
+     * replayed thinking blocks — either the official API itself or a proxy
+     * that forwards to it (GitHub Copilot, ZenMux, Cloudflare AI Gateway's
+     * `/anthropic` route, Google Vertex's `publishers/anthropic/…`).
+     * Downstream transforms strip stale cross-model thinking signatures on
+     * these endpoints so the signing proxy doesn't 400 with
+     * `Invalid signature in thinking block` (#4297), and adaptive-thinking
+     * models keep the interleaved-thinking beta on them (#6717). Known hosts
+     * are auto-detected from provider id and baseUrl; set this to mark an
+     * opaque signing proxy the URL list can't recognize. Superset of
+     * {@link ResolvedAnthropicCompat.officialEndpoint}.
+     */
+    signingEndpoint?: boolean;
 }
 /**
  * Compatibility settings for Bedrock Converse prompt caching. Cache pricing is
@@ -597,17 +611,6 @@ export type ResolvedAnthropicCompat = Required<AnthropicCompat> & {
      * env headers, and cache-TTL shaping without per-request URL parsing.
      */
     officialEndpoint: boolean;
-    /**
-     * The configured endpoint enforces Anthropic's signature protocol on
-     * replayed thinking blocks — either the official API itself or a proxy
-     * that forwards to it (GitHub Copilot, ZenMux, Cloudflare AI Gateway's
-     * `/anthropic` route, Google Vertex's `publishers/anthropic/…`).
-     * Downstream transforms strip stale cross-model thinking signatures on
-     * these endpoints so the signing proxy doesn't 400 with
-     * `Invalid signature in thinking block` (#4297). Superset of
-     * {@link officialEndpoint}.
-     */
-    signingEndpoint: boolean;
 };
 /**
  * Compatibility settings for the devin-agent (Codeium Cascade) API. Cascade

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "17.1.4",
+	"version": "17.1.6",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,12 +34,12 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.1",
-		"@oh-my-pi/pi-utils": "17.1.4",
+		"@oh-my-pi/pi-utils": "17.1.6",
 		"arktype": "2.2.3",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "17.1.4",
+		"@oh-my-pi/pi-ai": "17.1.6",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/anthropic.ts

@@ -4,7 +4,7 @@
  * defaults come from provider ids, strict host checks, and model-id
  * classification, with explicit spec overrides assigned on top.
  */
-import { modelMatchesHost } from "../hosts";
+import { hostMatchesUrl, modelMatchesHost } from "../hosts";
 import {
 	hasOpus47ApiRestrictions,
 	isAnthropicFableOrMythosModel,
@@ -96,6 +96,23 @@ function isAzureAnthropicRoute(baseUrl?: string): boolean {
 	return baseUrl !== undefined && AZURE_ANTHROPIC_URL_MARKER.test(baseUrl);
 }
 
+/**
+ * Known non-official URLs that enforce Anthropic thinking signatures on replay.
+ *
+ * Runtime routing calls this with the effective URL because a model's resolved
+ * compat can be stale after Foundry or a provider base-URL override reroutes it.
+ */
+export function isAnthropicSigningProxyUrl(baseUrl?: string): boolean {
+	return (
+		hostMatchesUrl(baseUrl, "githubCopilot") ||
+		hostMatchesUrl(baseUrl, "zenmux") ||
+		isCloudflareAnthropicGateway(baseUrl) ||
+		isVertexAnthropicRoute(baseUrl) ||
+		isBedrockAnthropicRoute(baseUrl) ||
+		isAzureAnthropicRoute(baseUrl)
+	);
+}
+
 /** Build the resolved anthropic-messages compat record for a model spec. */
 export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): ResolvedAnthropicCompat {
 	const baseUrl = spec.baseUrl;
@@ -113,11 +130,8 @@ export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): Res
 	// (issue #4192).
 	const isZenmux = modelMatchesHost(spec, "zenmux");
 	const requiresThinkingEnabled = modelMatchesHost(spec, "moonshotNative") && matchesKimiMandatoryThinkingModel(spec);
-	const isVertex = isVertexAnthropicRoute(baseUrl);
-	const isBedrock = isBedrockAnthropicRoute(baseUrl);
 	const isAzure = isAzureAnthropicRoute(baseUrl);
-	const signingEndpoint =
-		official || isCopilot || isZenmux || isCloudflareAnthropicGateway(baseUrl) || isVertex || isBedrock || isAzure;
+	const signingEndpoint = official || isCopilot || isZenmux || isAnthropicSigningProxyUrl(baseUrl);
 	const compat: ResolvedAnthropicCompat = {
 		officialEndpoint: official,
 		signingEndpoint,

package/src/model-cache.ts

@@ -9,8 +9,10 @@ import type { Api, Model, ModelSpec } from "./types";
 // Rows persist ModelSpec JSON (sparse `compat`, never the resolved record);
 // the model manager rebuilds via `buildModel` on load. Request headers are
 // intentionally omitted: arbitrary provider-defined header names can carry
-// credentials. v11 invalidates rows that may persist derived computer-use
-// support without provenance; v10 deletes rows that may contain persisted
+// credentials. v12 invalidates Kimi Code rows carrying the blanket
+// maxTokens: 32000 that predate per-family output caps (k3/k3-256k -> 131072,
+// kimi-for-coding[-highspeed] -> 32768, #6711); v11 invalidates rows that may
+// persist derived computer-use
 // headers and records which model ids lost headers or cannot be rebuilt.
 // v9 invalidated Kimi Code rows predating live effort and protocol metadata;
 // v8 invalidated Codex discovery rows predating provider-native V2 compaction
@@ -20,7 +22,7 @@ import type { Api, Model, ModelSpec } from "./types";
 // retired unknown-limit sentinels (222222/8888); v5 invalidated rows predating
 // effort-tier variant collapsing (raw `-low`/`-high`/`-thinking` member ids);
 // v4 dropped the pre-efforts ThinkingConfig shape.
-const CACHE_SCHEMA_VERSION = 11;
+const CACHE_SCHEMA_VERSION = 12;
 const HEADER_RESTORE_VERSION = 1;
 
 interface CacheRow {