@oh-my-pi/pi-catalog 18.8.6 → 18.8.7

package/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [18.8.7] - 2026-10-09
+
+### Added
+
+- Added built-in CoralBricks support with `/login`, live model discovery, per-model reasoning levels and off controls, and bundled offline fallbacks. ([#14146](https://github.com/can1357/oh-my-pi/pull/14146) by [@ryan-brosas](https://github.com/ryan-brosas))
+- Added `gen:models --provider <id>` to update one provider without changing other providers' catalog snapshots. ([#14146](https://github.com/can1357/oh-my-pi/pull/14146) by [@ryan-brosas](https://github.com/ryan-brosas))
+
+### Fixed
+
+- Fixed Claude Haiku 5.5 thinking Off to request explicitly disabled thinking on every host serving its adaptive thinking ([#14996](https://github.com/can1357/oh-my-pi/pull/14996) by [@bse-ai](https://github.com/bse-ai)).
+
 ## [18.8.6] - 2026-10-08
 
 ### Added

package/dist/types/compat/auth-ids.d.ts

@@ -1,4 +1,4 @@
 /** Every provider with an `auth/<id>.kdl` policy. */
-export type AuthProviderId = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
+export type AuthProviderId = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
 /** Providers whose policy declares a `login` flow (the `/login` roster). */
-export type LoginProviderId = "abliteration" | "aiand" | "alibaba-coding-plan" | "alibaba-token-plan" | "anthropic" | "baseten" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";
+export type LoginProviderId = "abliteration" | "aiand" | "alibaba-coding-plan" | "alibaba-token-plan" | "anthropic" | "baseten" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "helmcode" | "huggingface" | "kagi" | "kilo" | "kimi-code" | "litellm" | "llama.cpp" | "lm-studio" | "meta" | "minimax-code" | "minimax-code-cn" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai-codex" | "openai-codex-device" | "opencode-go" | "opencode-zen" | "openrouter" | "parallel" | "perplexity" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stencil" | "stepfun" | "synthetic" | "tavily" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zai-coding-plan" | "zenmux" | "zhipu-coding-plan";

package/dist/types/compat/provider-ids.d.ts

@@ -1,2 +1,2 @@
 /** Every chat-model provider with a catalog entry (`default-model` in its provider rules). */
-export type KnownProvider = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kilo" | "kimi-code" | "litellm" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "opencode-go" | "opencode-zen" | "openrouter" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stepfun" | "synthetic" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zenmux" | "zhipu-coding-plan";
+export type KnownProvider = "abliteration" | "aiand" | "aimlapi" | "alibaba-coding-plan" | "alibaba-token-plan" | "amazon-bedrock" | "anthropic" | "apple" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coralbricks" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "factory-droid" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "helmcode" | "huggingface" | "kilo" | "kimi-code" | "litellm" | "lm-studio" | "local" | "meta" | "minimax" | "minimax-code" | "minimax-code-cn" | "mistral" | "moonshot" | "muse-code" | "nanogpt" | "novita" | "nvidia" | "ollama" | "ollama-cloud" | "openai" | "openai-codex" | "opencode-go" | "opencode-zen" | "openrouter" | "qianfan" | "qwen-portal" | "sakana" | "siliconflow" | "siliconflow-cn" | "singularityapi-dev" | "singularityapi-tech" | "snowflake" | "stepfun" | "synthetic" | "together" | "typesafe" | "umans" | "venice" | "vercel-ai-gateway" | "vllm" | "wafer-serverless" | "web" | "xai" | "xai-oauth" | "xiaomi" | "xiaomi-token-plan-ams" | "xiaomi-token-plan-cn" | "xiaomi-token-plan-sgp" | "yolo-auto" | "zai" | "zenmux" | "zhipu-coding-plan";

package/dist/types/model-manager.d.ts

@@ -35,6 +35,22 @@ export interface ModelManagerOptions<TApi extends Api = Api, TModelsDevPayload =
     cacheTtlMs?: number;
     /** When true, a successful dynamic fetch is the complete provider catalog and prunes static-only models. */
     dynamicModelsAuthoritative?: boolean;
+    /**
+     * When true, a provider's dynamic rows are the deployment's whole truth
+     * for input modality: a live text-only row strips the bundled reference's
+     * image support instead of OR-merging it back. Declared by providers whose
+     * capability flags are authoritative and whose endpoints reject
+     * unsupported content (`github-copilot`, `deepinfra`, `coralbricks`).
+     */
+    dynamicInputAuthoritative?: boolean;
+    /** When true, live reasoning capability replaces the fallback, including an explicit false. */
+    dynamicReasoningAuthoritative?: boolean;
+    /**
+     * When true, a live per-million price replaces the fallback even when it is
+     * an explicit `0` (a free model or a dropped cache-write charge). Without it,
+     * a zero discovery price is treated as unknown and the bundled rate is kept.
+     */
+    dynamicCostAuthoritative?: boolean;
     /**
      * When true, a fresh cache never satisfies an online-eligible refresh: the
      * dynamic fetch always runs (an explicit `"offline"` strategy is still

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

@@ -154,6 +154,19 @@ export interface DeepinfraModelManagerConfig {
  * model availability at runtime is still gated on credentials by the registry.
  */
 export declare function deepinfraModelManagerOptions(config?: DeepinfraModelManagerConfig): ModelManagerOptions<"openai-completions">;
+export declare const CORALBRICKS_BASE_URL = "https://inference.coralbricks.ai/v1";
+/** CoralBricks OpenAI-compatible discovery configuration. */
+export interface CoralbricksModelManagerConfig {
+    apiKey?: string;
+    baseUrl?: string;
+    fetch?: FetchImpl;
+}
+/**
+ * Builds CoralBricks' model-discovery manager. `/v1/models` is key-protected
+ * (401 without a bearer key), so a keyless config serves only the bundled
+ * reviewed seed rows; with a key, live rows are authoritative over the bundle.
+ */
+export declare function coralbricksModelManagerOptions(config?: CoralbricksModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export interface XaiModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/dist/types/types.d.ts

@@ -192,6 +192,8 @@ export interface OpenAICompat {
     supportsMultipleSystemMessages?: boolean;
     /** Whether the provider supports `reasoning_effort`. Default: auto-detected from URL. */
     supportsReasoningEffort?: boolean;
+    /** Do not infer a thinking dial when discovery supplies no explicit thinking configuration. */
+    trustExplicitThinkingOnly?: boolean;
     /** Optional mapping from pi-ai reasoning levels to provider/model-specific `reasoning_effort` values. */
     reasoningEffortMap?: Partial<Record<Effort, string>>;
     /** Whether the provider supports `stream_options: { include_usage: true }` for token usage in streaming responses. Default: true. */
@@ -646,6 +648,8 @@ export interface AnthropicCompat {
  * deliberately not used to infer these request-shape capabilities.
  */
 export interface BedrockCompat {
+    /** Explicit disabled-thinking wire form; unset preserves the provider's existing behavior. */
+    disabledThinking?: AnthropicCompat["disabledThinking"];
     /** Whether this endpoint accepts no checkpoints, automatic caching, or explicit cachePoint blocks. */
     promptCacheMode?: "none" | "automatic" | "explicit";
     /** Whether this wire may revise already-streamed text (`stream-revision` axis). Unassigned: append-only. */
@@ -684,6 +688,8 @@ export interface BedrockCompat {
 }
 /** Fully-resolved Bedrock Converse prompt-cache capabilities, materialized once by `buildModel`. */
 export interface ResolvedBedrockCompat {
+    /** See {@link BedrockCompat.disabledThinking}. */
+    disabledThinking?: BedrockCompat["disabledThinking"];
     promptCacheMode: NonNullable<BedrockCompat["promptCacheMode"]>;
     /** See {@link BedrockCompat.streamRevision}. */
     streamRevision?: BedrockCompat["streamRevision"];
@@ -738,6 +744,7 @@ export interface ResolvedOpenAISharedCompat {
     supportsDeveloperRole: boolean;
     supportsStrictMode: boolean;
     supportsReasoningEffort: boolean;
+    trustExplicitThinkingOnly?: boolean;
     reasoningEffortMap: Partial<Record<Effort, string>>;
     supportsReasoningParams: boolean;
     supportsSamplingParams: boolean;
@@ -814,7 +821,7 @@ export interface ResolvedOpenAISharedCompat {
  * `buildModel`; request handlers read fields and never detect, resolve, or
  * allocate.
  */
-export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "trustExplicitThinkingOnly" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "statefulResponses" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "mistralReasoningContentParts" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "syntheticReasoningContentFallback" | "replayReasoningContent" | "qwenPreserveThinking" | "qwenTemplateReasoningEffort" | "requiresThinkingAsText" | "requiresMistralToolIds" | "requiresToolResultName" | "requiresAssistantAfterToolResult" | "requiresAssistantContentForToolCalls" | "stripDeepseekSpecialTokens" | "streamMarkupHealingPattern" | "reasoningDeltasMayBeCumulative" | "emptyLengthFinishIsContextError" | "usesOpenAIToolCallIdLimit" | "promptCacheSessionHeader" | "supportsPromptCacheBreakpoints" | "promptCacheBreakpointTtl" | "openRouterRouting" | "isOpenRouterHost" | "supportsStrictMode" | "supportsLongPromptCacheRetention" | "alwaysSendMaxTokens" | "clampOutputToModelMax" | "wireModelIdMode" | "vercelGatewayRouting" | "extraBody" | "toolStrictMode" | "toolSchemaFlavor" | "streamFirstEventTimeoutMs" | "streamIdleTimeoutMs" | "streamRevision" | "cacheControlFormat" | "thinkingKeep" | "strictResponsesPairing" | "supportsImageDetailOriginal" | "supportsConfigurationUpdate" | "supportsSteering" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
     vercelGatewayRouting?: OpenAICompat["vercelGatewayRouting"];
     extraBody?: OpenAICompat["extraBody"];
     cacheControlFormat?: OpenAICompat["cacheControlFormat"];

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.8.6",
+	"version": "18.8.7",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -93,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.8.6",
-		"@oh-my-pi/pi-utils": "18.8.6"
+		"@oh-my-pi/omptype": "18.8.7",
+		"@oh-my-pi/pi-utils": "18.8.7"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.8.6",
+		"@oh-my-pi/pi-ai": "18.8.7",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/auth-ids.ts

@@ -18,6 +18,7 @@ export type AuthProviderId =
 	| "cline-pass"
 	| "cloudflare-ai-gateway"
 	| "commandcode"
+	| "coralbricks"
 	| "coreweave"
 	| "cursor"
 	| "deepinfra"
@@ -110,6 +111,7 @@ export type LoginProviderId =
 	| "cline-pass"
 	| "cloudflare-ai-gateway"
 	| "commandcode"
+	| "coralbricks"
 	| "coreweave"
 	| "cursor"
 	| "deepinfra"

package/src/compat/axes.ts

@@ -229,7 +229,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"allow-anthropic-header-overrides": wire("allowAnthropicHeaderOverrides", ["anthropic"]),
 	"disable-adaptive-thinking": wire("disableAdaptiveThinking", ["anthropic"]),
 	"disable-strict-tools": wire("disableStrictTools", ["anthropic"]),
-	"disabled-thinking": wire("disabledThinking", ["anthropic"], "scalar", ["omit", "disabled", "adaptive"]),
+	"disabled-thinking": wire("disabledThinking", ["anthropic", "bedrock"], "scalar", ["omit", "disabled", "adaptive"]),
 	"effort-beta": wire("effortBeta", ["anthropic"]),
 	"escape-builtin-tool-names": wire("escapeBuiltinToolNames", ["anthropic"]),
 	"fast-mode": wire("fastMode", ["anthropic"]),
@@ -261,7 +261,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	// ── wire: devin-agent ──
 	"model-router": wire("modelRouter", ["devin"]),
 	"supports-parallel-tool-calls": wire("supportsParallelToolCalls", ["devin"]),
-	"trust-explicit-thinking-only": wire("trustExplicitThinkingOnly", ["devin"]),
+	"trust-explicit-thinking-only": wire("trustExplicitThinkingOnly", [...OAI, "devin"]),
 
 	// ── wire: google APIs ──
 	"antigravity-claude-tool-mode": wire("antigravityClaudeToolMode", ["google"]),

package/src/compat/provider-ids.ts

@@ -18,6 +18,7 @@ export type KnownProvider =
 	| "cline-pass"
 	| "cloudflare-ai-gateway"
 	| "commandcode"
+	| "coralbricks"
 	| "coreweave"
 	| "cursor"
 	| "deepinfra"

package/src/compat/resolve.ts

@@ -503,6 +503,7 @@ function detectOpenAICompat(
 		supportsDeveloperRole: isOpenAIHost || isAzureHost,
 		supportsMultipleSystemMessages: supportsMultipleSystemMessagesDefault,
 		supportsReasoningEffort: !isGrok && !d.isXiaomiMimo && (!(d.isZai || d.isZhipu) || supportsZaiReasoningEffort),
+		trustExplicitThinkingOnly: undefined,
 		// API-conditional: this completions-only Copilot exclusion cannot be a
 		// provider rule without changing Copilot Responses rows.
 		supportsReasoningParams: provider !== "github-copilot",
@@ -732,6 +733,7 @@ function resolveOpenAIResponsesPolicy(
 			hostMatchesUrl(baseUrl, "openrouter") ||
 			hostMatchesUrl(baseUrl, "deepseekFamily"),
 		supportsReasoningEffort: !isXaiHost,
+		trustExplicitThinkingOnly: undefined,
 		supportsLongPromptCacheRetention: isOpenAIUrl,
 		supportsPromptCacheBreakpoints,
 		promptCacheBreakpointTtl: supportsPromptCacheBreakpoints ? "30m" : undefined,

package/src/compat/rules/auth/_order.kdl

@@ -10,4 +10,4 @@ login-order "openai-codex" "anthropic" "zai" "zai-coding-plan" "kimi-code" "open
 	"qianfan" "venice" "siliconflow" "siliconflow-cn" "synthetic" "nanogpt" "wafer-serverless" "coreweave" \
 	"vercel-ai-gateway" "cloudflare-ai-gateway" "snowflake" "litellm" "kilo" "zenmux" "opencode-zen" "opencode-go" \
 	"yolo-auto" "tavily" "kagi" "exa" "parallel" "typesafe" "ollama" "ollama-cloud" "lm-studio" "llama.cpp" "vllm" \
-	"gmi-cloud" "stencil" "singularityapi-dev" "singularityapi-tech"
+	"gmi-cloud" "stencil" "singularityapi-dev" "singularityapi-tech" "coralbricks"

package/src/compat/rules/auth/coralbricks.kdl

@@ -0,0 +1,13 @@
+// `/v1/models` is key-protected — it answers `401 missing_api_key` without a
+// bearer key and `401 invalid_api_key` to a rejected one (verified
+// 2026-10-02) — so login validates the pasted key there. The probe bills
+// nothing, unlike a chat-completions ping against a prepaid balance.
+auth "coralbricks" {
+	name "CoralBricks"
+	login "api-key" {
+		auth-url "https://www.coralbricks.ai/api-keys"
+		instructions "Create or copy your API key from the CoralBricks dashboard (keys look like cb_...)"
+		prompt "Paste your CoralBricks API key" placeholder="cb_..."
+		validate "models-endpoint" label="CoralBricks" url="https://inference.coralbricks.ai/v1/models"
+	}
+}

package/src/compat/rules/classes/anthropic.kdl

@@ -308,8 +308,12 @@ class "anthropic" {
 		family "haiku" {
 			thinking-mode "budget"
 			// Haiku 5.5 is the first adaptive Haiku: `budget_tokens` is a 400.
+			// An omitted `thinking` is adaptive, but unlike Sonnet/Opus 5.5 it
+			// accepts `disabled` at effort `high` or below, so Off is explicit.
+			// https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-haiku-5-5#use-effort-to-control-thinking
 			revision ">=5.5 <6" {
 				thinking-mode "anthropic-adaptive"
+				disabled-thinking "disabled"
 			}
 		}
 	}

package/src/compat/rules/providers/coralbricks.kdl

@@ -0,0 +1,62 @@
+// CoralBricks inference gateway deployment contract
+// (https://www.coralbricks.ai/docs.md; catalog snapshot 2026-10-08).
+//
+// Coral serves open-weight models behind an OpenAI-compatible wire at
+// https://inference.coralbricks.ai/v1 with a single bearer API key. The
+// key-protected `GET /v1/models` publishes each row's context window,
+// per-million pricing, and capability flags, and Coral documents those
+// flags as authoritative (unsupported content answers
+// `400 unsupported_content_type` rather than being dropped), so live
+// discovery is the deployment truth and same-id rows on other hosts must
+// not backfill capabilities or limits into this provider's rows.
+provider "coralbricks" {
+	default-model "glm-5.3-fast"
+	// Official docs authenticate via `CORAL_API_KEY`
+	// (https://www.coralbricks.ai/docs.md#authentication); `CORALBRICKS_API_KEY`
+	// is the alias the Coral pi provider extension shipped with, kept as a
+	// compatibility fallback in precedence order.
+	env "CORAL_API_KEY" "CORALBRICKS_API_KEY"
+	dynamic-models-authoritative #true
+	skip-cross-provider-reference-fills #true
+	discovery label="CoralBricks"
+
+	// Reviewed snapshot of the public catalog
+	// (https://www.coralbricks.ai/api/public/models). Coral retired GLM 5.3
+	// Flash on 2026-10-07 and renamed the remaining slugs (`glm-5.3-fast`,
+	// `deepseek-v4.1-flash-fast`). The public rows list input/output rates
+	// only; cache-write rates come from Coral's published pricing and cached
+	// reads are $0 on every model. `/v1/models` needs a key, so generation
+	// without credentials bundles these rows; successful runtime discovery
+	// replaces them.
+	seed api="openai-completions" base-url="https://inference.coralbricks.ai/v1" bundle="always" {
+		model "glm-5.3-fast" name="GLM 5.3" {
+			reasoning #true
+			input "text"
+			cost input=1.12 output=4.4 cache-read=0 cache-write=1.68
+			limits context=1048576 max-tokens=131072
+		}
+		model "deepseek-v4.1-flash-fast" name="DeepSeek V4.1 Flash" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.3 output=1.2 cache-read=0 cache-write=0.09
+			limits context=1048576 max-tokens=131072
+		}
+	}
+
+	// DeepSeek V4.1 Flash is the one CoralBricks SKU that advertises a genuine
+	// off value (`none` in its reasoning-effort vocabulary, per the reasoning
+	// table at https://www.coralbricks.ai/docs.md#reasoning), so a
+	// thinking-off request disables reasoning instead of clamping to the
+	// lowest effort like the GLM SKUs.
+	models "deepseek-v4.1-flash-fast" {
+		reasoning-disable-mode "none-effort"
+	}
+	// Gateway request dialect, matching Coral's plain Chat Completions
+	// surface ("unchanged from what your existing SDK expects"): output caps
+	// ride `max_tokens`, system prompts ride the `system` role, and no
+	// `store` field is sent. Coral accepts unknown fields, so acceptance is
+	// not evidence the semantics are honored (same call as charm-hyper).
+	supports-developer-role #false
+	supports-store #false
+	max-tokens-field "max_tokens"
+}