@oh-my-pi/pi-catalog 18.2.11 → 18.3.1

package/CHANGELOG.md

@@ -2,6 +2,27 @@
 
 ## [Unreleased]
 
+## [18.3.1] - 2026-09-25
+
+### Added
+
+- Added compatibility flags for models that stop output at the context-window limit and models that support steering, including defaults for Claude 4.5+ and GPT-6+ models.
+
+### Fixed
+
+- Fixed forced-tool requests for Claude Opus 5.5 so tool selection falls back gracefully when necessary.
+- Added the provider-advertised `xhigh` reasoning level for Yolo-Auto Qwen3.8 models.
+- Fixed Devin Fusion pairing requests failing with `no API providers are available` by routing them through the lead model with the lead's limits and pricing; pairings without an available lead are no longer listed ([#13000](https://github.com/can1357/oh-my-pi/pull/13000) by [@DarkPhilosophy](https://github.com/DarkPhilosophy)).
+
+## [18.3.0] - 2026-09-24
+
+### Added
+
+- Added compaction support for the `claude-mythos-preview` model.
+- Added provider and authentication rule definitions for on-device Apple Foundation Models.
+- Added per-account discovery of Codex cyber access entitlements.
+- Added `org-scoped-identity` and `oauth-token-env` authentication rules for providers.
+
 ## [18.2.11] - 2026-09-23
 
 ### Added

package/README.md

@@ -1,6 +1,6 @@
 # @oh-my-pi/pi-catalog
 
-Model catalog for [oh-my-pi](https://github.com/can1357/oh-my-pi): bundled model database, provider discovery, model identity, classification, and equivalence.
+Model catalog for [omp](https://github.com/can1357/oh-my-pi): bundled model database, provider discovery, model identity, classification, and equivalence.
 
 ## What's inside
 

package/THIRD-PARTY-NOTICES.txt

@@ -78,8 +78,8 @@ below.
 uutils coreutils (https://github.com/uutils/coreutils), MIT
 --------------------------------------------------------------------------------
 Covers: base32, base64, basename, cat, cksum (shared checksum machinery),
-b2sum, md5sum, sha1sum, sha224sum, sha256sum, sha384sum, sha512sum, comm, cut,
-date, dirname, head, hostname, ln, ls, mkdir, mktemp, mv, nproc, paste,
+b2sum, md5sum, sha1sum, sha224sum, sha256sum, sha384sum, sha512sum, comm, cp,
+cut, date, dirname, head, hostname, ln, ls, mkdir, mktemp, mv, nproc, paste,
 printenv, readlink, realpath, rm, seq, sort, stat, tac, tail, tee, touch, tr,
 truncate, uname, uniq, wc, whoami, yes.
 
@@ -293,7 +293,7 @@ MIT License
 Copyright (c) 2026 Sander Land
   (measured tokenizer vocabulary data, reconstruction model, and reference
   implementation: https://github.com/sanderland/ctok)
-Copyright (c) 2026 Can Bölük and the Oh My Pi contributors
+Copyright (c) 2026 Can Bölük and the omp contributors
 Copyright (c) 2026 Stencil Labs, Inc.
   (Rust implementation and the compact binary vocabulary encoding in
   crates/pi-natives/src/utok/claude)

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" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "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" | "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" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "exa" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "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" | "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" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google-antigravity" | "google-gemini-cli" | "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" | "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/fallback-credit.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Model ids on which a refusal's Anthropic fallback-credit token may be
+ * redeemed, authored on the `fallback-credit-targets` catalog axis. Empty when
+ * no rule assigns targets (the model cannot redeem credit on any fallback).
+ */
+export declare function fallbackCreditTargets(model: Model): readonly string[];

package/dist/types/compat/output-limits.d.ts

@@ -1,3 +1,8 @@
 import type { Model } from "../types.js";
+/**
+ * Whether the host truncates generation at the context window rather than
+ * rejecting prompt + output cap overflow; the output fitter skips these models.
+ */
+export declare function stopsOutputAtContextWindow(model: Model): boolean;
 /** Whether discovery and transport policy allow preserving a caller's output cap. */
 export declare function supportsOutputTokenLimit(model: Model): boolean;

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" | "azure" | "baseten" | "bedrock-mantle" | "cerebras" | "charm-hyper" | "cline-pass" | "cloudflare-ai-gateway" | "commandcode" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "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" | "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" | "coreweave" | "cursor" | "deepinfra" | "deepseek" | "devin" | "firepass" | "fireworks" | "github-copilot" | "gitlab-duo" | "gitlab-duo-agent" | "gmi-cloud" | "google" | "google-antigravity" | "google-gemini-cli" | "google-vertex" | "groq" | "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" | "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/compat/server-side-fallback.d.ts

@@ -0,0 +1,7 @@
+import type { Model } from "../types.js";
+/**
+ * Ordered model ids for Anthropic's server-side refusal `fallbacks` chain,
+ * authored on the `server-side-fallback-models` catalog axis. Empty when no
+ * rule assigns a chain (the model or host does not support the feature).
+ */
+export declare function serverSideFallbackModels(model: Model): readonly string[];

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

@@ -530,6 +530,10 @@ export interface CompiledAuthProvider {
         hook: string;
     };
     allowsMissingApiKey?: boolean;
+    /** Qualify credential and usage-report identity by org when an email may have multiple subscriptions. */
+    orgScopedIdentity?: boolean;
+    /** Environment variables carrying this provider's own OAuth bearer, excluding borrowed API-key aliases. */
+    oauthTokenEnv?: string[];
     /** APIs whose provider transport resolves credentials without a stored account. */
     nativeAuthApis?: string[];
     available?: boolean;

package/dist/types/discovery/cursor-proto.d.ts

@@ -1,5 +1,5 @@
 /**
- * Cursor agent protocol declarations used by Oh My Pi.
+ * Cursor agent protocol declarations used by omp.
  *
  * Each declaration retains only fields consumed by the client or its protocol tests.
  */

package/dist/types/discovery/devin-proto.d.ts

@@ -1,5 +1,5 @@
 /**
- * Devin protocol declarations used by Oh My Pi.
+ * Devin protocol declarations used by omp.
  *
  * Each declaration retains only fields consumed by the client or its protocol tests.
  */

package/dist/types/types.d.ts

@@ -2,7 +2,7 @@ import type { ModelIdentity } from "./compat/types.js";
 import type { Effort } from "./effort.js";
 export type { FetchImpl } from "@oh-my-pi/pi-utils";
 export type { KnownProvider } from "./provider-models/descriptors.js";
-export type KnownApi = "openai-completions" | "openai-responses" | "openrouter" | "openai-codex-responses" | "azure-openai-responses" | "anthropic-messages" | "bedrock-converse-stream" | "google-generative-ai" | "google-gemini-cli" | "google-vertex" | "ollama-chat" | "cursor-agent" | "gitlab-duo-agent" | "devin-agent";
+export type KnownApi = "openai-completions" | "openai-responses" | "openrouter" | "openai-codex-responses" | "azure-openai-responses" | "anthropic-messages" | "bedrock-converse-stream" | "google-generative-ai" | "google-gemini-cli" | "google-vertex" | "ollama-chat" | "cursor-agent" | "gitlab-duo-agent" | "devin-agent" | "apple-foundation-models";
 export type Api = KnownApi | (string & {});
 /** Catalog kinds used to isolate role-specific runners from session chat models. */
 export declare const MODEL_KINDS: readonly ["chat", "tiny", "image", "tts", "stt", "search", "judge", "embedding", "rerank", "video"];
@@ -401,6 +401,13 @@ export interface OpenAICompat {
      * sent as the top-level `reasoning.effort`.
      */
     supportsConfigurationUpdate?: boolean;
+    /**
+     * Whether the Responses WebSocket accepts `response.steer`, which queues user
+     * input into the in-flight response (GPT-6 family). Default: rule-detected.
+     * Set `false` for proxies that reject the event; steering then waits for the
+     * next request boundary.
+     */
+    supportsSteering?: boolean;
     /** Whether streamed reasoning deltas for the same field may repeat the full cumulative text snapshot. Default: false. */
     reasoningDeltasMayBeCumulative?: boolean;
     /** Strip leaked DeepSeek chat-template special tokens from visible content deltas. Default: auto-detected. */
@@ -441,17 +448,17 @@ export interface AnthropicCompat {
     /** Whether thinking requests may include `context_management` and its beta header. Default: true. */
     supportsContextManagement?: boolean;
     /**
-     * Whether the model lineage supports Anthropic server-side compaction
-     * (`compact-2026-01-12`: the `compact_20260112` edit and replayed
-     * `compaction` blocks). Rule-owned per model line; the beta covers the
-     * adaptive-thinking generation onward and rejects older lines. Default: false.
+     * Whether the model and host support Anthropic on-demand compaction
+     * (`compact-2026-09-04` requests and signed replay). Enabled on Opus 4.6+,
+     * Sonnet 4.6+, Fable/Mythos 5+ on supported hosts. Default: false.
      */
     supportsServerCompaction?: boolean;
     /**
      * Whether the model is served by the first-party Anthropic provider (its
      * default route is the official API). Rule-owned on the provider; the
      * compaction transport pairs it with a per-request effective-URL check
-     * because reroutes leave it stale-true. Default: false.
+     * because reroutes leave it stale-true. Vertex is selected by its provider
+     * contract instead. Default: false.
      */
     firstPartyProvider?: boolean;
     /**
@@ -727,7 +734,7 @@ 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" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "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" | "stripImageInput" | "thinkingLoopGuard" | "whenThinking">> & {
+export type ResolvedOpenAICompat = ResolvedOpenAISharedCompat & Required<Omit<OpenAICompat, "supportsDeveloperRole" | "supportsReasoningEffort" | "reasoningEffortMap" | "supportsReasoningParams" | "supportsReasoningSummary" | "supportsSamplingParams" | "supportsPenaltyAndStopParams" | "thinkingFormat" | "kimiApiFormat" | "reasoningDisableMode" | "omitReasoningEffort" | "includeEncryptedReasoning" | "filterReasoningHistory" | "disableReasoningOnForcedToolChoice" | "disableReasoningOnToolChoice" | "disableReasoningWithTools" | "supportsToolChoice" | "supportsForcedToolChoice" | "supportsNamedToolChoice" | "reasoningContentField" | "requiresReasoningContentForToolCalls" | "requiresReasoningContentForAllAssistantTurns" | "allowsSyntheticReasoningContentForToolCalls" | "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"];
@@ -761,6 +768,11 @@ export interface ResolvedOpenAIResponsesCompat extends ResolvedOpenAISharedCompa
      * the item type with 400.
      */
     supportsConfigurationUpdate: boolean;
+    /**
+     * Whether the WebSocket transport may send `response.steer` to deliver user
+     * input into the in-flight response. Rule-owned: GPT-6 family.
+     */
+    supportsSteering: boolean;
     /** Inject the `# Juice: 0 !important` developer item when reasoning is forced off (gpt-5.6+). */
     requiresReasoningOffJuiceInstruction: boolean;
     /**
@@ -953,6 +965,15 @@ export interface ModelCost extends TokenCost {
  * `"deepseek-v3"`; Kimi K2 through K3 share `"kimi-k2"`.
  */
 export type ModelTokenizer = "claude-v3" | "claude-v47" | "claude-v5" | "claude-v5-sonnet" | "qwen3" | "deepseek-v3" | "kimi-k2" | "glm5";
+/** One account's discovered entitlements on a model; see {@link Model.accountAccess}. */
+export interface ModelAccountAccess {
+    /**
+     * Codex `available_access_programs.cyber`: cyber access programs this account
+     * may request on the model (`standard`, `daybreak_blue`, `daybreak_red`).
+     * Absent when the backend reported no program metadata.
+     */
+    cyberPrograms?: readonly string[];
+}
 export interface Model<TApi extends Api = Api> {
     id: string;
     /** Role-specific runner capability; omitted for ordinary chat models. */
@@ -1046,6 +1067,15 @@ export interface Model<TApi extends Api = Api> {
      * their single wire id) and on bundled snapshots that predate discovery.
      */
     cursorMaxModeRoutes?: Readonly<Record<string, boolean>>;
+    /**
+     * Per-account availability recorded by multi-account discovery: provider
+     * account id (Codex: ChatGPT `chatgpt_account_id`) → that account's
+     * entitlements on this model. An account appears only when its own catalog
+     * lists the model, so credential selection can route account-gated models
+     * (e.g. `gpt-daybreak-blue-latest`) straight to eligible accounts. Absent on
+     * bundled/config rows and on single-account discovery.
+     */
+    accountAccess?: Readonly<Record<string, ModelAccountAccess>>;
     cost: ModelCost;
     /** Premium Copilot requests charged per user-initiated request (defaults to 1). */
     premiumMultiplier?: number;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "18.2.11",
+	"version": "18.3.1",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"keywords": [
 		"ai",
@@ -14,7 +14,10 @@
 		"url": "https://github.com/can1357/oh-my-pi/issues"
 	},
 	"license": "MIT",
-	"author": "Stencil Labs, Inc.",
+	"author": {
+		"name": "Stencil Labs, Inc.",
+		"url": "https://stencil.so"
+	},
 	"repository": {
 		"type": "git",
 		"url": "git+https://github.com/can1357/oh-my-pi.git",
@@ -90,12 +93,12 @@
 		"gen:proto": "bun scripts/generate-protocols.ts"
 	},
 	"dependencies": {
-		"@oh-my-pi/omptype": "18.2.11",
-		"@oh-my-pi/pi-utils": "18.2.11"
+		"@oh-my-pi/omptype": "18.3.1",
+		"@oh-my-pi/pi-utils": "18.3.1"
 	},
 	"devDependencies": {
 		"@bgotink/kdl": "0.4.0",
-		"@oh-my-pi/pi-ai": "18.2.11",
+		"@oh-my-pi/pi-ai": "18.3.1",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/auth-ids.ts

@@ -9,6 +9,7 @@ export type AuthProviderId =
 	| "alibaba-token-plan"
 	| "amazon-bedrock"
 	| "anthropic"
+	| "apple"
 	| "azure"
 	| "baseten"
 	| "bedrock-mantle"

package/src/compat/axes.ts

@@ -139,6 +139,7 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 	"requires-reasoning-off-juice-instruction": wire("requiresReasoningOffJuiceInstruction", ["openai-responses"]),
 	"supports-all-turns-reasoning-context": wire("supportsAllTurnsReasoningContext", ["openai-responses"]),
 	"supports-configuration-update": wire("supportsConfigurationUpdate", ["openai-responses"]),
+	"supports-steering": wire("supportsSteering", ["openai-responses"]),
 	"strip-deepseek-special-tokens": wire("stripDeepseekSpecialTokens", OAI),
 	"stream-markup-healing-pattern": wire("streamMarkupHealingPattern", OAI, "scalar", [
 		"kimi",
@@ -303,6 +304,18 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 		shape: "scalar",
 		values: [true, false],
 	},
+	/**
+	 * The host accepts a prompt plus `max_tokens` beyond the context window and
+	 * ends generation at the window (Anthropic `model_context_window_exceeded`)
+	 * instead of rejecting the request, so callers must not lower the output cap
+	 * to fit the window.
+	 */
+	"stops-output-at-context-window": {
+		key: "stopsOutputAtContextWindow",
+		set: "catalog",
+		shape: "scalar",
+		values: [true, false],
+	},
 	"clamp-context-override": { key: "clampContextOverride", set: "catalog", shape: "scalar" },
 	"context-promotion-target": { key: "contextPromotionTarget", set: "catalog", shape: "scalar" },
 	"context-window-floor": { key: "contextWindowFloor", set: "catalog", shape: "scalar" },
@@ -335,6 +348,18 @@ export const AXES: Readonly<Record<string, AxisDef>> = {
 		shape: "scalar",
 	},
 	"supports-assistant-prefill": { key: "supportsAssistantPrefill", set: "catalog", shape: "scalar" },
+	/**
+	 * Ordered Anthropic model ids forwarded as the server-side `fallbacks`
+	 * chain when the user opts in. Each must appear in the requested model's
+	 * `allowed_fallback_models` (GET /v1/models/{id}); anything else is a 400.
+	 */
+	"server-side-fallback-models": { key: "serverSideFallbackModels", set: "catalog", shape: "array" },
+	/**
+	 * Anthropic model ids a refusal's `fallback_credit_token` may be redeemed
+	 * on (the refused model's permitted fallback targets). Unordered; a retry
+	 * on any other model cannot redeem the credit.
+	 */
+	"fallback-credit-targets": { key: "fallbackCreditTargets", set: "catalog", shape: "array" },
 	priority: { key: "priority", set: "catalog", shape: "scalar" },
 	"service-tier-cost": { key: "serviceTierCost", set: "catalog", shape: "object" },
 	"time-based-cost": { key: "timeBased", set: "catalog", shape: "object" },

package/src/compat/fallback-credit.ts

@@ -0,0 +1,13 @@
+import type { Model } from "../types";
+import { resolveCatalogPolicy } from "./catalog-policy";
+
+/**
+ * Model ids on which a refusal's Anthropic fallback-credit token may be
+ * redeemed, authored on the `fallback-credit-targets` catalog axis. Empty when
+ * no rule assigns targets (the model cannot redeem credit on any fallback).
+ */
+export function fallbackCreditTargets(model: Model): readonly string[] {
+	const value = resolveCatalogPolicy(model).fallbackCreditTargets;
+	if (!Array.isArray(value)) return [];
+	return value.filter((entry): entry is string => typeof entry === "string");
+}

package/src/compat/output-limits.ts

@@ -1,6 +1,14 @@
 import type { Model } from "../types";
 import { resolveCatalogPolicy } from "./catalog-policy";
 
+/**
+ * Whether the host truncates generation at the context window rather than
+ * rejecting prompt + output cap overflow; the output fitter skips these models.
+ */
+export function stopsOutputAtContextWindow(model: Model): boolean {
+	return resolveCatalogPolicy(model).stopsOutputAtContextWindow === true;
+}
+
 /** Whether discovery and transport policy allow preserving a caller's output cap. */
 export function supportsOutputTokenLimit(model: Model): boolean {
 	const policy = resolveCatalogPolicy(model);

package/src/compat/provider-ids.ts

@@ -9,6 +9,7 @@ export type KnownProvider =
 	| "alibaba-token-plan"
 	| "amazon-bedrock"
 	| "anthropic"
+	| "apple"
 	| "azure"
 	| "baseten"
 	| "bedrock-mantle"

package/src/compat/resolve.ts

@@ -726,6 +726,7 @@ function resolveOpenAIResponsesPolicy(
 		supportsReasoningSummary: !isXaiHost,
 		supportsAllTurnsReasoningContext: false,
 		supportsConfigurationUpdate: false,
+		supportsSteering: false,
 		requiresReasoningOffJuiceInstruction: false,
 		stripImageInput: false,
 		thinkingLoopGuard: undefined,
@@ -832,6 +833,7 @@ function pickResponsesOnly(compat: ResolvedOpenAIResponsesCompat): ResponsesOnly
 		supportsObfuscationOptOut: compat.supportsObfuscationOptOut,
 		supportsAllTurnsReasoningContext: compat.supportsAllTurnsReasoningContext,
 		supportsConfigurationUpdate: compat.supportsConfigurationUpdate,
+		supportsSteering: compat.supportsSteering,
 		officialEndpoint: compat.officialEndpoint,
 		harmonyLeakMitigation: compat.harmonyLeakMitigation,
 		cacheControlFormat: compat.cacheControlFormat,

package/src/compat/rules/README.md

@@ -346,12 +346,16 @@ auth "anthropic" {
     expiry "jwt-or-never"                        // session-JWT expiry policy
     result "api-key"                             // OAuth login persists only credentials.access as a plain API key
     allows-missing-api-key #true
+    org-scoped-identity #true                   // qualify credential/report identity by org (one email can have multiple subscriptions)
+    oauth-token-env "PROVIDER_OAUTH_TOKEN"      // dedicated OAuth bearer env vars, excluding borrowed API-key aliases in provider env
     native-auth-api "bedrock-converse-stream"     // provider transport resolves auth; scan plans pin this API without secrets
     available #false
     show-in-login-list #false
 }
 ```
 
+`org-scoped-identity #true` keeps credentials and usage reports for different organizations separate even when they share an email; absent or `#false` uses ordinary account identity. `oauth-token-env` takes one or more ordered, non-empty env names carrying this provider's own OAuth bearer. When present, availability ignores other provider-env aliases, and usage probes accept only stored OAuth credentials or the first set bearer from that list; absent providers keep the ordinary API-key env behavior.
+
 Login kinds:
 
 - `login "api-key" { auth-url "…"; instructions "…"; prompt "…" placeholder="…"; empty-fallback "…"; normalize "strip-bearer"; validate … }` — `validate "chat-completions" base-url= model= tolerate-model-denied= max-tokens-field= max-tokens=`, `validate "anthropic-messages" base-url= model=`, or `validate "models-endpoint" url= base-url-env= headers-hook=`; all accept `label=` (error-message label, defaults to `name`) and `optional=#true` (only auth failures reject).

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

@@ -1,5 +1,6 @@
 auth "anthropic" {
 	name "Anthropic (Claude Pro/Max)"
+	org-scoped-identity #true
 	// Foundry mode optionally switches Anthropic auth to enterprise gateway credentials.
 	env hook="anthropic-foundry"
 	login "oauth-code" {

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

@@ -0,0 +1,4 @@
+auth "apple" {
+	name "Apple Foundation Models (on-device)"
+	allows-missing-api-key #true
+}

package/src/compat/rules/auth/openai-codex.kdl

@@ -1,5 +1,6 @@
 auth "openai-codex" {
 	name "ChatGPT Plus/Pro (Codex Subscription)"
+	org-scoped-identity #true
 	login "oauth-code" {
 		client-id "app_EMoamEEZ73f0CkXaXp7hrann"
 		authorize-url "https://auth.openai.com/oauth/authorize"

package/src/compat/rules/auth/xai-oauth.kdl

@@ -1,6 +1,7 @@
 // Device authorization and token refresh adapted from NousResearch/hermes-agent (MIT).
 auth "xai-oauth" {
     name "xAI Grok OAuth (SuperGrok or X Premium+)"
+    oauth-token-env "XAI_OAUTH_TOKEN"
     login "device-code" {
         client-id "b1a00492-073a-47ea-816f-4c329264a828"
         scopes "openid" "profile" "email" "offline_access" "grok-cli:access" "api:access"

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

@@ -7,28 +7,58 @@ class "anthropic" {
 			thinking-prefix-binding #true
 		}
 	}
-	// Server-side compaction (compact-2026-01-12) is a lineage capability: the
-	// documented model list is the adaptive-thinking generation onward (Opus
-	// 4.6+, Sonnet 4.6+, Fable/Mythos 5+). Whether a host delivers the beta is
-	// decided at request time from the effective endpoint, not here.
-	family "fable" {
-		revision ">=5" {
-			supports-server-compaction #true
+	// On-demand compaction (compact-2026-09-04) is available on the Claude
+	// API and Vertex, not Amazon Bedrock or unverified gateway deployments.
+	// Foundry / Claude Platform on AWS are first-party Anthropic routes whose
+	// effective URLs are checked at request time.
+	on "anthropic" "google-vertex" {
+		family "fable" {
+			revision ">=5" {
+				supports-server-compaction #true
+			}
 		}
-	}
-	family "mythos" {
-		revision ">=5" {
+		family "mythos" {
+			revision ">=5" {
+				supports-server-compaction #true
+			}
+		}
+		family "opus" {
+			revision ">=4.6" {
+				supports-server-compaction #true
+			}
+		}
+		family "sonnet" {
+			revision ">=4.6" {
+				supports-server-compaction #true
+			}
+		}
+		models "claude-mythos-preview" {
 			supports-server-compaction #true
 		}
 	}
-	family "opus" {
-		revision ">=4.6" {
-			supports-server-compaction #true
+	// Sonnet 4.5 and newer end generation with `model_context_window_exceeded`
+	// when prompt + max_tokens overruns the window, instead of a 400, so the
+	// full output cap can be requested without sizing the prompt. Verified for
+	// the Claude API only:
+	// https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#model-context-window-exceeded
+	on "anthropic" {
+		revision ">=4.5" {
+			stops-output-at-context-window #true
 		}
 	}
-	family "sonnet" {
-		revision ">=4.6" {
-			supports-server-compaction #true
+	// Opus 5.5 rejects forced tool selection the way Fable/Mythos do — a
+	// `tool_choice` of `tool` or `any` fails the request with 400
+	// `invalid_request_error`: `tool_choice: type "tool" and "any" are not
+	// supported for this model`. The transport downgrades the selector to
+	// `auto` so the tool stays offered instead of the turn dying.
+	//
+	// Upper-bounded on purpose: separator-collapsed ids parse their revision
+	// as a whole number (`claude-opus-45` is Opus 4.5, revision 45), so an
+	// open `>=5.5` would silently catch them, and a later Opus generation has
+	// to be re-verified against the API before it inherits the restriction.
+	family "opus" {
+		revision ">=5.5 <6" {
+			supports-forced-tool-choice #false
 		}
 	}
 	on "anthropic" "cloudflare-ai-gateway" "google-vertex" {

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

@@ -37,8 +37,11 @@ class "openai" {
 	revision "=5.6" {
 		delegation-bias "gated"
 	}
+	// GPT-6 accepts `response.steer` over the Responses WebSocket: user input
+	// joins the in-flight response instead of waiting for the next request.
 	revision ">=6" {
 		delegation-bias "restrained"
+		supports-steering #true
 	}
 	// GPT-6 Astra changes reasoning effort mid-conversation through a
 	// `configuration_update` input item so the request-level effort (and the

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

@@ -58,6 +58,21 @@ provider "anthropic" {
 		revision ">=3.7 <4.6" {
 			thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 		}
+		// Server-side refusal fallback chain (opt-in via
+		// `providers.anthropic.serverSideFallback`). Targets must be in the
+		// model's `allowed_fallback_models` on GET /v1/models/{id}; Fable 5 and
+		// 5.1 publish ["claude-opus-4-8", "claude-opus-5"] (2026-09-24).
+		// `claude-opus-5-5` is rejected with a 400 (#13059).
+		family "fable" {
+			server-side-fallback-models "claude-opus-5"
+			// Fallback-credit redemption targets: the refused model's permitted
+			// fallback targets. Fable 5 and 5.1 → Opus 4.8 and Opus 5
+			// (https://platform.claude.com/docs/en/build-with-claude/fallback-credit).
+			fallback-credit-targets "claude-opus-4-8" "claude-opus-5"
+		}
+		family "mythos" {
+			server-side-fallback-models "claude-opus-5"
+		}
 	}
 	// residue: taxonomy ranks and exact globs do not isolate these models.
 	models "*20240229" "claude-3-7-sonnet-20250219" {

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

@@ -0,0 +1,8 @@
+// Apple Foundation Models: the on-device system language model (macOS 27+,
+// Apple silicon), bridged in-process by pi-natives. Models are discovered at
+// runtime from the model's availability; nothing is bundled.
+
+provider "apple" {
+	default-model "on-device"
+	allow-unauthenticated #true
+}

package/src/compat/rules/providers/yolo-auto.kdl

@@ -37,7 +37,7 @@ provider "yolo-auto" {
 			cost input=0 output=0 cache-read=0 cache-write=0
 			limits context=262144 max-tokens=131072
 			thinking-mode "effort"
-			thinking-efforts "minimal" "low" "medium" "high"
+			thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 			supports-developer-role #false
 			supports-store #false
 			supports-reasoning-effort #true
@@ -50,7 +50,7 @@ provider "yolo-auto" {
 			cost input=0 output=0 cache-read=0 cache-write=0
 			limits context=262144 max-tokens=131072
 			thinking-mode "effort"
-			thinking-efforts "minimal" "low" "medium" "high"
+			thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 			supports-developer-role #false
 			supports-store #false
 			supports-reasoning-effort #true
@@ -61,6 +61,12 @@ provider "yolo-auto" {
 
 	supports-developer-role #false
 	supports-store #false
+	// Live /v1/models advertises `thinking: minimal..xhigh` for the Qwen3.8
+	// rows; the cascade rule carries it into the regenerated bundle, whose
+	// thinking re-bake ignores seed-level ladders.
+	models "qwen3.8-flash" "yolo" "qwen3.8-27b" {
+		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
+	}
 	models "deepseek-flash-v4" {
 		disable-reasoning-on-tool-choice #true
 		max-tokens-field "max_completion_tokens"