@revealui/ai 0.6.2 → 0.6.4

package/README.md

@@ -178,7 +178,7 @@ long-running operations stop instead of just being ignored.
 
 Some MCP servers need LLM capabilities without bundling a provider. The
 spec lets servers issue `sampling/createMessage` requests — the *client*
-runs the inference, keeps cost + context control, and returns the result.
+runs the inference, keeps cost + context control, and returns the result (`packages/ai/src/tools/mcp-sampling.ts:207`).
 On the Ubuntu reference stack, that means servers get LLM access via the
 developer's local Canonical Inference Snap — no cloud round-trip required.
 
@@ -365,8 +365,8 @@ The memory system uses deep cloning to ensure immutability and prevent data corr
 3. **AgentContextManager Level**
    - `getContext()`: Returns value from cloned context (no additional cloning)
    - `getAllContext()`: Returns cloned context from WorkingMemory
-   - `setContext()`: Validates then sets (cloning happens in WorkingMemory)
-   - `updateContext()`: Validates then updates (cloning happens in WorkingMemory)
+   - `setContext()`: Validates then sets (cloning happens in WorkingMemory) (`packages/ai/src/memory/agent/context-manager.ts:84`)
+   - `updateContext()`: Validates then updates (cloning happens in WorkingMemory) (`packages/ai/src/memory/agent/context-manager.ts:131`)
 
 ### Performance Implications
 
@@ -450,7 +450,7 @@ manager.setContext('user', { name: 'John', age: 30 })
 
 ### Size Limits
 
-The system enforces limits to prevent performance issues:
+The system enforces limits (`packages/ai/src/memory/utils/validation.ts:28-39`) to prevent performance issues:
 
 - **Max Context Keys**: 10,000 keys
 - **Max Context Size**: ~10MB (approximate)

package/dist/a2a/card.d.ts

@@ -8,6 +8,15 @@
  */
 import type { AgentDefinition } from '@revealui/contracts';
 import { type A2AAgentCard } from '@revealui/contracts';
+/**
+ * Built-in platform agents defer model selection to the inference provider the
+ * account configures on the API Keys page. The dispatch runtime resolves the
+ * real model from that provider, so the definition itself names no vendor. An
+ * empty value means "use the configured provider's model". Claude and other
+ * model ids stay valid where an operator wires a compatible provider; they are
+ * no longer the hardcoded default.
+ */
+export declare const PROVIDER_RESOLVED_MODEL = "";
 declare class AgentCardRegistry {
     private readonly defs;
     constructor();

package/dist/a2a/card.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"card.d.ts","sourceRoot":"","sources":["../../src/a2a/card.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAC3D,OAAO,EAAE,KAAK,YAAY,EAAyB,MAAM,qBAAqB,CAAC;AAwI/E,cAAM,iBAAiB;IACrB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAsC;;IAO3D,QAAQ,CAAC,GAAG,EAAE,eAAe,GAAG,IAAI;IAIpC,UAAU,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;IAIpC,MAAM,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,EAAE,IAAI,GAAG,SAAS,CAAC,CAAC,GAAG,OAAO;IAOzF,MAAM,CAAC,OAAO,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS;IAIpD,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,YAAY,GAAG,IAAI;IAM9D,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,YAAY,EAAE;IAI1C,QAAQ,IAAI,eAAe,EAAE;IAI7B,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;CAG9B;AAED,qDAAqD;AACrD,eAAO,MAAM,iBAAiB,mBAA0B,CAAC;AAEzD,YAAY,EAAE,eAAe,EAAE,CAAC"}
+{"version":3,"file":"card.d.ts","sourceRoot":"","sources":["../../src/a2a/card.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,qBAAqB,CAAC;AAC3D,OAAO,EAAE,KAAK,YAAY,EAAyB,MAAM,qBAAqB,CAAC;AAM/E;;;;;;;GAOG;AACH,eAAO,MAAM,uBAAuB,KAAK,CAAC;AAoI1C,cAAM,iBAAiB;IACrB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAsC;;IAO3D,QAAQ,CAAC,GAAG,EAAE,eAAe,GAAG,IAAI;IAIpC,UAAU,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;IAIpC,MAAM,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,IAAI,CAAC,eAAe,EAAE,IAAI,GAAG,SAAS,CAAC,CAAC,GAAG,OAAO;IAOzF,MAAM,CAAC,OAAO,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS;IAIpD,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,YAAY,GAAG,IAAI;IAM9D,SAAS,CAAC,OAAO,EAAE,MAAM,GAAG,YAAY,EAAE;IAI1C,QAAQ,IAAI,eAAe,EAAE;IAI7B,GAAG,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO;CAG9B;AAED,qDAAqD;AACrD,eAAO,MAAM,iBAAiB,mBAA0B,CAAC;AAEzD,YAAY,EAAE,eAAe,EAAE,CAAC"}

package/dist/a2a/card.js

@@ -10,6 +10,15 @@ import { agentDefinitionToCard } from '@revealui/contracts';
 // =============================================================================
 // Built-in agent definitions
 // =============================================================================
+/**
+ * Built-in platform agents defer model selection to the inference provider the
+ * account configures on the API Keys page. The dispatch runtime resolves the
+ * real model from that provider, so the definition itself names no vendor. An
+ * empty value means "use the configured provider's model". Claude and other
+ * model ids stay valid where an operator wires a compatible provider; they are
+ * no longer the hardcoded default.
+ */
+export const PROVIDER_RESOLVED_MODEL = '';
 /** RevealUI platform meta-agent  -  "The Creator" */
 const THE_CREATOR_DEF = {
     id: 'revealui-creator',
@@ -18,7 +27,7 @@ const THE_CREATOR_DEF = {
     description: 'The RevealUI platform agent. Scaffolds new AI agents, manages agent lifecycles, ' +
         'orchestrates multi-agent workflows, and acts as the primary interface for AI capabilities ' +
         'on the RevealUI platform.',
-    model: 'claude-opus-4-6',
+    model: PROVIDER_RESOLVED_MODEL,
     systemPrompt: 'You are The Creator, the meta-agent for RevealUI. You design, configure, and deploy ' +
         'purpose-built AI agents for RevealUI users. You have access to agent scaffolding tools, ' +
         'the RevealUI admin, and the billing system.',
@@ -74,7 +83,7 @@ const TICKET_AGENT_DEF = {
     name: 'Ticket Agent',
     description: 'Handles support tickets, resolves user issues, and escalates when needed. ' +
         'Uses the RevealUI admin to create and update tickets.',
-    model: 'claude-sonnet-4-6',
+    model: PROVIDER_RESOLVED_MODEL,
     systemPrompt: 'You are the RevealUI Ticket Agent. You help users resolve issues by creating tickets, ' +
         'searching for solutions, and escalating complex problems to the support team.',
     tools: [

package/dist/llm/cache-utils.d.ts

@@ -53,7 +53,7 @@ export declare function estimateCacheSavings(inputTokens: number, cacheHitRate:
  *
  * @example
  * ```ts
- * const response = await client.chat(messages, { enableCache: true })
+ * const response = await client.chat(messages, { cacheHint: true })
  * const stats = formatCacheStats(response.usage)
  * console.log(stats)
  * // "Cache: 45% read (2,500 tokens), 10% created (500 tokens)"
@@ -83,7 +83,7 @@ export declare function shouldCache(content: string, minTokens?: number): boolea
  *   ],
  * })
  *
- * const response = await client.chat(conversation, { enableCache: true })
+ * const response = await client.chat(conversation, { cacheHint: true })
  * ```
  */
 export declare function createCachedConversation(config: {
@@ -92,31 +92,17 @@ export declare function createCachedConversation(config: {
     messages: Message[];
 }): Message[];
 /**
- * Pricing information for Anthropic Claude models (as of 2024)
- * Prices are per million tokens
+ * @deprecated Use `MODEL_PRICING` from `./token-counter.js`. Retained as a thin derived
+ * view (the Anthropic-with-cache subset) so existing importers keep one source of truth.
  */
 export declare const ANTHROPIC_PRICING: {
-    readonly 'claude-3-5-sonnet-20241022': {
-        readonly input: 3;
-        readonly output: 15;
-        readonly cacheWrite: 3.75;
-        readonly cacheRead: 0.3;
-    };
-    readonly 'claude-3-5-haiku-20241022': {
-        readonly input: 1;
-        readonly output: 5;
-        readonly cacheWrite: 1.25;
-        readonly cacheRead: 0.1;
-    };
-    readonly 'claude-3-opus-20240229': {
-        readonly input: 15;
-        readonly output: 75;
-        readonly cacheWrite: 18.75;
-        readonly cacheRead: 1.5;
-    };
+    readonly 'claude-3-5-sonnet-20241022': import("./token-counter.js").ModelPricing;
+    readonly 'claude-3-5-haiku-20241022': import("./token-counter.js").ModelPricing;
+    readonly 'claude-3-opus-20240229': import("./token-counter.js").ModelPricing;
 };
 /**
- * Calculate actual cost of a request with caching
+ * Calculate actual cost of a request with caching, using the unified MODEL_PRICING table.
+ * Unknown models price at zero (matching `estimateCost`).
  *
  * @example
  * ```ts
@@ -128,12 +114,12 @@ export declare const ANTHROPIC_PRICING: {
  *   cacheReadTokens: 5000,
  * })
  *
- * console.log(`Request cost: $${cost.toFixed(4)}`)
+ * console.log(`Request cost: $${cost.total.toFixed(4)}`)
  * console.log(`Savings vs no cache: $${cost.savings.toFixed(4)}`)
  * ```
  */
 export declare function calculateCacheCost(usage: {
-    model: keyof typeof ANTHROPIC_PRICING;
+    model: string;
     promptTokens: number;
     completionTokens: number;
     cacheCreationTokens?: number;

package/dist/llm/cache-utils.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cache-utils.d.ts","sourceRoot":"","sources":["../../src/llm/cache-utils.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAEnD;;;GAGG;AACH,wBAAgB,SAAS,CAAC,OAAO,EAAE,OAAO,GAAG,OAAO,CAKnD;AAED;;;;;;;;;GASG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAK9D;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,MAAM,EACnB,YAAY,EAAE,MAAM,EACpB,qBAAqB,EAAE,MAAM,GAC5B,MAAM,CAeR;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE;IACtC,YAAY,EAAE,MAAM,CAAC;IACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG,MAAM,GAAG,IAAI,CAqBhB;AAED;;;GAGG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,SAAO,GAAG,OAAO,CAItE;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,wBAAwB,CAAC,MAAM,EAAE;IAC/C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,EAAE,OAAO,EAAE,CAAC;CACrB,GAAG,OAAO,EAAE,CAwBZ;AAED;;;GAGG;AACH,eAAO,MAAM,iBAAiB;;;;;;;;;;;;;;;;;;;CAmBpB,CAAC;AAEX;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE;IACxC,KAAK,EAAE,MAAM,OAAO,iBAAiB,CAAC;IACtC,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAwBxE"}
+{"version":3,"file":"cache-utils.d.ts","sourceRoot":"","sources":["../../src/llm/cache-utils.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAGnD;;;GAGG;AACH,wBAAgB,SAAS,CAAC,OAAO,EAAE,OAAO,GAAG,OAAO,CAKnD;AAED;;;;;;;;;GASG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAK9D;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,oBAAoB,CAClC,WAAW,EAAE,MAAM,EACnB,YAAY,EAAE,MAAM,EACpB,qBAAqB,EAAE,MAAM,GAC5B,MAAM,CAeR;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE;IACtC,YAAY,EAAE,MAAM,CAAC;IACrB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG,MAAM,GAAG,IAAI,CAqBhB;AAED;;;GAGG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,MAAM,EAAE,SAAS,SAAO,GAAG,OAAO,CAItE;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,wBAAwB,CAAC,MAAM,EAAE;IAC/C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,QAAQ,EAAE,OAAO,EAAE,CAAC;CACrB,GAAG,OAAO,EAAE,CAwBZ;AAWD;;;GAGG;AACH,eAAO,MAAM,iBAAiB;;;;CAIpB,CAAC;AAEX;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAgB,kBAAkB,CAAC,KAAK,EAAE;IACxC,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,gBAAgB,EAAE,MAAM,CAAC;IACzB,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B,GAAG;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CA6BxE"}

package/dist/llm/cache-utils.js

@@ -1,4 +1,5 @@
 /* console-allowed */
+import { MODEL_PRICING } from './token-counter.js';
 /**
  * Mark a message for caching
  * Use this for system prompts, tool definitions, or large context that repeats
@@ -6,7 +7,7 @@
 export function withCache(message) {
     return {
         ...message,
-        cacheControl: { type: 'ephemeral' },
+        cache: true,
     };
 }
 /**
@@ -57,7 +58,7 @@ export function estimateCacheSavings(inputTokens, cacheHitRate, cachedTokenPerce
  *
  * @example
  * ```ts
- * const response = await client.chat(messages, { enableCache: true })
+ * const response = await client.chat(messages, { cacheHint: true })
  * const stats = formatCacheStats(response.usage)
  * console.log(stats)
  * // "Cache: 45% read (2,500 tokens), 10% created (500 tokens)"
@@ -102,7 +103,7 @@ export function shouldCache(content, minTokens = 1024) {
  *   ],
  * })
  *
- * const response = await client.chat(conversation, { enableCache: true })
+ * const response = await client.chat(conversation, { cacheHint: true })
  * ```
  */
 export function createCachedConversation(config) {
@@ -118,7 +119,7 @@ export function createCachedConversation(config) {
             result.push({
                 role: 'system',
                 content: doc,
-                ...(isLast ? { cacheControl: { type: 'ephemeral' } } : {}),
+                ...(isLast ? { cache: true } : {}),
             });
         });
     }
@@ -126,32 +127,26 @@ export function createCachedConversation(config) {
     result.push(...config.messages);
     return result;
 }
+/** Look up a model's pricing, asserting presence (used only for known-present keys). */
+function requirePricing(model) {
+    const pricing = MODEL_PRICING[model];
+    if (!pricing) {
+        throw new Error(`No pricing entry for model: ${model}`);
+    }
+    return pricing;
+}
 /**
- * Pricing information for Anthropic Claude models (as of 2024)
- * Prices are per million tokens
+ * @deprecated Use `MODEL_PRICING` from `./token-counter.js`. Retained as a thin derived
+ * view (the Anthropic-with-cache subset) so existing importers keep one source of truth.
  */
 export const ANTHROPIC_PRICING = {
-    'claude-3-5-sonnet-20241022': {
-        input: 3.0,
-        output: 15.0,
-        cacheWrite: 3.75, // 25% markup for cache creation
-        cacheRead: 0.3, // 90% discount for cache hits
-    },
-    'claude-3-5-haiku-20241022': {
-        input: 1.0,
-        output: 5.0,
-        cacheWrite: 1.25,
-        cacheRead: 0.1,
-    },
-    'claude-3-opus-20240229': {
-        input: 15.0,
-        output: 75.0,
-        cacheWrite: 18.75,
-        cacheRead: 1.5,
-    },
+    'claude-3-5-sonnet-20241022': requirePricing('claude-3-5-sonnet-20241022'),
+    'claude-3-5-haiku-20241022': requirePricing('claude-3-5-haiku-20241022'),
+    'claude-3-opus-20240229': requirePricing('claude-3-opus-20240229'),
 };
 /**
- * Calculate actual cost of a request with caching
+ * Calculate actual cost of a request with caching, using the unified MODEL_PRICING table.
+ * Unknown models price at zero (matching `estimateCost`).
  *
  * @example
  * ```ts
@@ -163,12 +158,17 @@ export const ANTHROPIC_PRICING = {
  *   cacheReadTokens: 5000,
  * })
  *
- * console.log(`Request cost: $${cost.toFixed(4)}`)
+ * console.log(`Request cost: $${cost.total.toFixed(4)}`)
  * console.log(`Savings vs no cache: $${cost.savings.toFixed(4)}`)
  * ```
  */
 export function calculateCacheCost(usage) {
-    const pricing = ANTHROPIC_PRICING[usage.model];
+    const pricing = MODEL_PRICING[usage.model] ?? {
+        input: 0,
+        output: 0,
+        cacheWrite: 0,
+        cacheRead: 0,
+    };
     const uncachedTokens = usage.promptTokens - (usage.cacheCreationTokens || 0) - (usage.cacheReadTokens || 0);
     const costs = {
         input: (uncachedTokens / 1_000_000) * pricing.input,

package/dist/llm/providers/base.d.ts

@@ -1,7 +1,11 @@
 /**
  * LLM Provider Base Interface
  *
- * Abstract interface for all LLM providers (OpenAI, Anthropic, etc.)
+ * The agnostic Reasoner port (ADR 2026-06-25). Providers are negotiated by
+ * `capabilities()`, never by provider-name string. The control vocabulary is neutral
+ * and capability-gated (`effort`, `cache`/`cacheHint`); genuinely unportable knobs ride
+ * the namespaced opaque `providerOptions` channel; cross-provider usage (e.g. cache token
+ * stats) stays neutral in `LLMResponse.usage`.
  */
 /**
  * A plain text content part  -  used in multipart messages.
@@ -36,10 +40,11 @@ export interface Message {
     name?: string;
     toolCalls?: ToolCall[];
     toolCallId?: string;
-    /** Anthropic prompt caching - marks content for caching (5min TTL, 90% cost reduction) */
-    cacheControl?: {
-        type: 'ephemeral';
-    };
+    /**
+     * Neutral cache breakpoint hint (gated by the `promptCache` capability). An adapter
+     * that supports prompt caching maps this to its native cache-control; others ignore it.
+     */
+    cache?: boolean;
 }
 export interface ToolCall {
     id: string;
@@ -58,7 +63,11 @@ export interface LLMResponse {
         promptTokens: number;
         completionTokens: number;
         totalTokens: number;
-        /** Anthropic cache stats */
+        /**
+         * Cross-provider cache usage (neutral; gated by `promptCache`). Populated by any
+         * caching provider  -  Anthropic cache tokens, OpenAI `cached_tokens`, Gemini
+         * `cachedContentTokenCount`. Read by the cost path; stays in neutral `usage`.
+         */
         cacheCreationTokens?: number;
         cacheReadTokens?: number;
     };
@@ -81,22 +90,66 @@ export interface LLMProviderConfig {
     maxTokens?: number;
 }
 /**
- * Base interface for LLM providers
+ * Reasoning-depth hint (neutral, capability-gated by `reasoningEffort`).
+ *
+ * This is a *reasoning-depth* axis, NOT output entropy or length  -  adapters must never
+ * remap it to `temperature`/`maxTokens`. An adapter with a reasoning axis maps it to its
+ * native control (e.g. a thinking-token budget); an adapter without one advertises
+ * `reasoningEffort: false` and treats `effort` as a no-op.
  */
-export interface LLMProvider {
-    /**
-     * Chat completion
-     */
+export type ReasoningEffort = 'off' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh';
+/**
+ * Runtime-queryable capability record. The loop branches on these, never on a
+ * provider-name string. Conformance-tested (transport tier) so a provider cannot lie:
+ * advertise-true-but-not-wired fails the positive assertion; advertise-false-but-emits-a-
+ * native-field fails the inertness assertion.
+ */
+export interface ReasonerCapabilities {
+    /** Stable adapter id; also the only `providerOptions` namespace this adapter may read ('x-<tag>'). */
+    readonly providerTag: string;
+    readonly tools: boolean;
+    /** Implies `tools` (consistency invariant). */
+    readonly parallelToolCalls: boolean;
+    /** Implies the adapter accepts `ContentPart[]` message content. */
+    readonly vision: boolean;
+    readonly streaming: boolean;
+    readonly embeddings: boolean;
+    readonly reasoningEffort: boolean;
+    readonly promptCache: boolean;
+    readonly structuredOutput: boolean;
+    /** Max context window in tokens; undefined = unknown (loop budgets conservatively). */
+    readonly contextWindow?: number;
+}
+/**
+ * Namespaced opaque extension for genuinely unportable, request-shaped knobs.
+ *
+ * Core/loop code MUST pass this through and MUST NOT read any key  -  per-namespace schema
+ * lives in the owning adapter. A known-namespace registry (typed keys) rather than a bare
+ * `x-*` string convention, so a typo'd namespace fails loudly instead of silently no-opping.
+ */
+export interface ProviderOptions {
+    'x-anthropic'?: unknown;
+    'x-openai'?: unknown;
+    'x-google'?: unknown;
+}
+/**
+ * The Reasoner port  -  one swappable reasoning step plus the side capabilities
+ * (embeddings/streaming). The loop negotiates by `capabilities()`, never by name.
+ */
+export interface Reasoner {
+    /** Runtime-queryable, conformance-tested capability record. */
+    capabilities(): ReasonerCapabilities;
+    /** Chat completion. */
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
-    /**
-     * Generate embeddings
-     */
+    /** Generate embeddings. */
     embed(text: string | string[], options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;
-    /**
-     * Stream chat completion
-     */
+    /** Stream chat completion. */
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
 }
+/**
+ * @deprecated Use `Reasoner`. Alias retained one release cycle for external callers.
+ */
+export type LLMProvider = Reasoner;
 export interface LLMChatOptions {
     temperature?: number;
     maxTokens?: number;
@@ -107,14 +160,12 @@ export interface LLMChatOptions {
             name: string;
         };
     };
-    /** Enable prompt caching (Anthropic only) - caches system prompts and tools */
-    enableCache?: boolean;
-    /**
-     * Extended thinking token budget (Anthropic only).
-     * Enables Claude's internal reasoning before responding.
-     * Typical values: 512 (minimal) → 31999 (xhigh). 0 or undefined = disabled.
-     */
-    thinkingBudget?: number;
+    /** Neutral reasoning-depth hint (gated by `reasoningEffort`). No-op where unsupported. */
+    effort?: ReasoningEffort;
+    /** Neutral prompt-cache hint for the stable prefix  -  system + tools (gated by `promptCache`). */
+    cacheHint?: boolean;
+    /** Opaque, namespaced provider extension. Core never reads it; the adapter owns the schema. */
+    providerOptions?: ProviderOptions;
 }
 export interface LLMEmbedOptions {
     model?: string;
@@ -123,8 +174,10 @@ export interface LLMStreamOptions {
     temperature?: number;
     maxTokens?: number;
     tools?: ToolDefinition[];
-    /** Enable prompt caching (Anthropic only) - caches system prompts and tools */
-    enableCache?: boolean;
+    /** Neutral reasoning-depth hint (gated by `reasoningEffort`). No-op where unsupported. */
+    effort?: ReasoningEffort;
+    /** Opaque, namespaced provider extension. Core never reads it; the adapter owns the schema. */
+    providerOptions?: ProviderOptions;
 }
 export interface ToolDefinition {
     type: 'function';

package/dist/llm/providers/base.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"base.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/base.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE;QACT,sEAAsE;QACtE,GAAG,EAAE,MAAM,CAAC;QACZ,yDAAyD;QACzD,MAAM,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;KAClC,CAAC;CACH;AAED,8DAA8D;AAC9D,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC;AAE/C,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,GAAG,MAAM,CAAC;IAC/C;yFACqF;IACrF,OAAO,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;IAChC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,0FAA0F;IAC1F,YAAY,CAAC,EAAE;QAAE,IAAI,EAAE,WAAW,CAAA;KAAE,CAAC;CACtC;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC;IACnE,KAAK,CAAC,EAAE;QACN,YAAY,EAAE,MAAM,CAAC;QACrB,gBAAgB,EAAE,MAAM,CAAC;QACzB,WAAW,EAAE,MAAM,CAAC;QACpB,4BAA4B;QAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;CACH;AAED,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;GAEG;AACH,MAAM,WAAW,WAAW;IAC1B;;OAEG;IACH,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC,CAAC;IAE1E;;OAEG;IACH,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC,CAAC;IAE5F;;OAEG;IACH,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;CAClF;AAED,MAAM,WAAW,cAAc;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IAChF,+EAA+E;IAC/E,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;OAIG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,+EAA+E;IAC/E,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACrC,CAAC;CACH;AAED,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC"}
+{"version":3,"file":"base.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/base.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE;QACT,sEAAsE;QACtE,GAAG,EAAE,MAAM,CAAC;QACZ,yDAAyD;QACzD,MAAM,CAAC,EAAE,KAAK,GAAG,MAAM,GAAG,MAAM,CAAC;KAClC,CAAC;CACH;AAED,8DAA8D;AAC9D,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,SAAS,CAAC;AAE/C,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,GAAG,MAAM,CAAC;IAC/C;yFACqF;IACrF,OAAO,EAAE,MAAM,GAAG,WAAW,EAAE,CAAC;IAChC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;;OAGG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;CACjB;AAED,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,SAAS,EAAE,MAAM,CAAC;KACnB,CAAC;CACH;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;IACvB,YAAY,CAAC,EAAE,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC;IACnE,KAAK,CAAC,EAAE;QACN,YAAY,EAAE,MAAM,CAAC;QACrB,gBAAgB,EAAE,MAAM,CAAC;QACzB,WAAW,EAAE,MAAM,CAAC;QACpB;;;;WAIG;QACH,mBAAmB,CAAC,EAAE,MAAM,CAAC;QAC7B,eAAe,CAAC,EAAE,MAAM,CAAC;KAC1B,CAAC;CACH;AAED,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,OAAO,CAAC;IACd,SAAS,CAAC,EAAE,QAAQ,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,iBAAiB;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAEtF;;;;;GAKG;AACH,MAAM,WAAW,oBAAoB;IACnC,sGAAsG;IACtG,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC;IACxB,+CAA+C;IAC/C,QAAQ,CAAC,iBAAiB,EAAE,OAAO,CAAC;IACpC,mEAAmE;IACnE,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,UAAU,EAAE,OAAO,CAAC;IAC7B,QAAQ,CAAC,eAAe,EAAE,OAAO,CAAC;IAClC,QAAQ,CAAC,WAAW,EAAE,OAAO,CAAC;IAC9B,QAAQ,CAAC,gBAAgB,EAAE,OAAO,CAAC;IACnC,uFAAuF;IACvF,QAAQ,CAAC,aAAa,CAAC,EAAE,MAAM,CAAC;CACjC;AAED;;;;;;GAMG;AACH,MAAM,WAAW,eAAe;IAC9B,aAAa,CAAC,EAAE,OAAO,CAAC;IACxB,UAAU,CAAC,EAAE,OAAO,CAAC;IACrB,UAAU,CAAC,EAAE,OAAO,CAAC;CACtB;AAED;;;GAGG;AACH,MAAM,WAAW,QAAQ;IACvB,+DAA+D;IAC/D,YAAY,IAAI,oBAAoB,CAAC;IAErC,uBAAuB;IACvB,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC,CAAC;IAE1E,2BAA2B;IAC3B,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC,CAAC;IAE5F,8BAA8B;IAC9B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC,CAAC;CAClF;AAED;;GAEG;AACH,MAAM,MAAM,WAAW,GAAG,QAAQ,CAAC;AAEnC,MAAM,WAAW,cAAc;IAC7B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IAChF,0FAA0F;IAC1F,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB,mGAAmG;IACnG,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,+FAA+F;IAC/F,eAAe,CAAC,EAAE,eAAe,CAAC;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,gBAAgB;IAC/B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,cAAc,EAAE,CAAC;IACzB,0FAA0F;IAC1F,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB,+FAA+F;IAC/F,eAAe,CAAC,EAAE,eAAe,CAAC;CACnC;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,UAAU,CAAC;IACjB,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,WAAW,EAAE,MAAM,CAAC;QACpB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACrC,CAAC;CACH;AAED,MAAM,MAAM,YAAY,GAAG,MAAM,GAAG,QAAQ,GAAG,YAAY,GAAG,gBAAgB,CAAC"}

package/dist/llm/providers/base.js

@@ -1,6 +1,10 @@
 /**
  * LLM Provider Base Interface
  *
- * Abstract interface for all LLM providers (OpenAI, Anthropic, etc.)
+ * The agnostic Reasoner port (ADR 2026-06-25). Providers are negotiated by
+ * `capabilities()`, never by provider-name string. The control vocabulary is neutral
+ * and capability-gated (`effort`, `cache`/`cacheHint`); genuinely unportable knobs ride
+ * the namespaced opaque `providerOptions` channel; cross-provider usage (e.g. cache token
+ * stats) stays neutral in `LLMResponse.usage`.
  */
 export {};

package/dist/llm/providers/groq.d.ts

@@ -5,7 +5,7 @@
  * Free tier: 6,000 TPM / 500k TPD.
  * Sign up: console.groq.com
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface GroqProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
     apiKey: string;
     /** Defaults to https://api.groq.com/openai/v1 */
@@ -16,6 +16,7 @@ export interface GroqProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
 export declare class GroqProvider implements LLMProvider {
     private inner;
     constructor(config: GroqProviderConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
     embed(_text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;

package/dist/llm/providers/groq.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"groq.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/groq.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,kBAAmB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC3E,MAAM,EAAE,MAAM,CAAC;IACf,iDAAiD;IACjD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,qBAAa,YAAa,YAAW,WAAW;IAC9C,OAAO,CAAC,KAAK,CAAuB;gBAExB,MAAM,EAAE,kBAAkB;IAQtC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAIhF,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,QAAQ,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAG9F"}
+{"version":3,"file":"groq.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/groq.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,kBAAmB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC3E,MAAM,EAAE,MAAM,CAAC;IACf,iDAAiD;IACjD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,qBAAa,YAAa,YAAW,WAAW;IAC9C,OAAO,CAAC,KAAK,CAAuB;gBAExB,MAAM,EAAE,kBAAkB;IAQtC,YAAY,IAAI,oBAAoB;IAepC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAIhF,KAAK,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE,EAAE,QAAQ,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAG9F"}

package/dist/llm/providers/groq.js

@@ -15,6 +15,20 @@ export class GroqProvider {
             model: config.model ?? 'qwen/qwen3-32b',
         });
     }
+    capabilities() {
+        // Profile for the default model (qwen/qwen3-32b). Groq exposes no embeddings endpoint.
+        return {
+            providerTag: 'groq',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: false,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     chat(messages, options) {
         return this.inner.chat(messages, options);
     }

package/dist/llm/providers/inference-snaps.d.ts

@@ -24,7 +24,7 @@
  *
  * Docs: https://documentation.ubuntu.com/inference-snaps
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface InferenceSnapsProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
     apiKey?: string;
     /** Base URL of the inference-snaps service, e.g. http://localhost:9090/v1 */
@@ -39,6 +39,7 @@ export declare class InferenceSnapsProvider implements LLMProvider {
     private embedModel;
     private baseURL;
     constructor(config: InferenceSnapsProviderConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
     embed(text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;

package/dist/llm/providers/inference-snaps.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"inference-snaps.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/inference-snaps.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,4BAA6B,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IACrF,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6EAA6E;IAC7E,OAAO,EAAE,MAAM,CAAC;IAChB,+FAA+F;IAC/F,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,qEAAqE;IACrE,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,sBAAuB,YAAW,WAAW;IACxD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,4BAA4B;IAahD,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAI1E,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,QAAQ,CAAC,EAAE,eAAe,GACzB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAsBpC"}
+{"version":3,"file":"inference-snaps.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/inference-snaps.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,4BAA6B,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IACrF,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,6EAA6E;IAC7E,OAAO,EAAE,MAAM,CAAC;IAChB,+FAA+F;IAC/F,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,qEAAqE;IACrE,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,sBAAuB,YAAW,WAAW;IACxD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,4BAA4B;IAahD,YAAY,IAAI,oBAAoB;IAgBpC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAI1E,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,QAAQ,CAAC,EAAE,eAAe,GACzB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAsBpC"}

package/dist/llm/providers/inference-snaps.js

@@ -41,6 +41,21 @@ export class InferenceSnapsProvider {
             model: config.model ?? 'gemma3',
         });
     }
+    capabilities() {
+        // Profile for the default model (gemma3, text-only). Per-model capability
+        // negotiation (e.g. qwen-vl vision, deepseek-r1 reasoningEffort) is a follow-up.
+        return {
+            providerTag: 'inference-snaps',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     chat(messages, options) {
         return this.inner.chat(messages, options);
     }

package/dist/llm/providers/ollama.d.ts

@@ -5,7 +5,7 @@
  * No API key required. Zero cost, fully offline.
  * Install: https://ollama.com
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface OllamaProviderConfig extends Omit<LLMProviderConfig, 'apiKey'> {
     apiKey?: string;
     /** Defaults to http://localhost:11434/v1 */
@@ -20,6 +20,7 @@ export declare class OllamaProvider implements LLMProvider {
     private embedModel;
     private baseURL;
     constructor(config: OllamaProviderConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;
     embed(text: string | string[], _options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;

package/dist/llm/providers/ollama.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ollama.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACR,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,oBAAqB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC7E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gFAAgF;IAChF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iGAAiG;IACjG,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,cAAe,YAAW,WAAW;IAChD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,oBAAoB;IAaxC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAI1E,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,QAAQ,CAAC,EAAE,eAAe,GACzB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAsBpC"}
+{"version":3,"file":"ollama.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/ollama.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EACT,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EACrB,MAAM,WAAW,CAAC;AAGnB,MAAM,WAAW,oBAAqB,SAAQ,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC;IAC7E,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gFAAgF;IAChF,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iGAAiG;IACjG,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,qBAAa,cAAe,YAAW,WAAW;IAChD,OAAO,CAAC,KAAK,CAAuB;IACpC,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,oBAAoB;IAaxC,YAAY,IAAI,oBAAoB;IAepC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IAIzE,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAI1E,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,QAAQ,CAAC,EAAE,eAAe,GACzB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;CAsBpC"}

package/dist/llm/providers/ollama.js

@@ -22,6 +22,20 @@ export class OllamaProvider {
             model: config.model ?? 'gemma4:e2b',
         });
     }
+    capabilities() {
+        // Profile for the default model (gemma4:e2b, text-only).
+        return {
+            providerTag: 'ollama',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     chat(messages, options) {
         return this.inner.chat(messages, options);
     }

package/dist/llm/providers/openai-compat.d.ts

@@ -5,13 +5,14 @@
  * format. Used by: Ollama, Groq, Inference Snaps.
  * NOT for direct OpenAI usage  -  RevealUI uses open-source models only.
  */
-import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message } from './base.js';
+import type { Embedding, LLMChatOptions, LLMChunk, LLMEmbedOptions, LLMProvider, LLMProviderConfig, LLMResponse, LLMStreamOptions, Message, ReasonerCapabilities } from './base.js';
 export interface OpenAICompatConfig extends LLMProviderConfig {
 }
 export declare class OpenAICompatProvider implements LLMProvider {
     private config;
     private baseURL;
     constructor(config: OpenAICompatConfig);
+    capabilities(): ReasonerCapabilities;
     chat(messages: Message[], options?: LLMChatOptions): Promise<LLMResponse>;
     embed(text: string | string[], options?: LLMEmbedOptions): Promise<Embedding | Embedding[]>;
     stream(messages: Message[], options?: LLMStreamOptions): AsyncIterable<LLMChunk>;

package/dist/llm/providers/openai-compat.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"openai-compat.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/openai-compat.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EAET,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EAER,MAAM,WAAW,CAAC;AAEnB,MAAM,WAAW,kBAAmB,SAAQ,iBAAiB;CAAG;AAqDhE,qBAAa,oBAAqB,YAAW,WAAW;IACtD,OAAO,CAAC,MAAM,CAAqB;IACnC,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,kBAAkB;IAUhC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IA6EzE,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;IAyC5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAuFvF,OAAO,CAAC,cAAc;CA4BvB"}
+{"version":3,"file":"openai-compat.d.ts","sourceRoot":"","sources":["../../../src/llm/providers/openai-compat.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EACV,SAAS,EAET,cAAc,EACd,QAAQ,EACR,eAAe,EACf,WAAW,EACX,iBAAiB,EACjB,WAAW,EACX,gBAAgB,EAChB,OAAO,EACP,oBAAoB,EAErB,MAAM,WAAW,CAAC;AAEnB,MAAM,WAAW,kBAAmB,SAAQ,iBAAiB;CAAG;AAqDhE,qBAAa,oBAAqB,YAAW,WAAW;IACtD,OAAO,CAAC,MAAM,CAAqB;IACnC,OAAO,CAAC,OAAO,CAAS;gBAEZ,MAAM,EAAE,kBAAkB;IAUtC,YAAY,IAAI,oBAAoB;IAgB9B,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC;IA6EzE,KAAK,CACT,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,SAAS,GAAG,SAAS,EAAE,CAAC;IAyC5B,MAAM,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE,gBAAgB,GAAG,aAAa,CAAC,QAAQ,CAAC;IAuFvF,OAAO,CAAC,cAAc;CA4BvB"}

package/dist/llm/providers/openai-compat.js

@@ -22,7 +22,7 @@ const asRecord = (value) => {
 };
 const isFunctionToolCall = (call) => {
     const record = asRecord(call);
-    if (!record || record.type !== 'function' || typeof record.id !== 'string') {
+    if (record?.type !== 'function' || typeof record.id !== 'string') {
         return false;
     }
     const fn = asRecord(record.function);
@@ -38,6 +38,21 @@ export class OpenAICompatProvider {
         }
         this.baseURL = config.baseURL;
     }
+    capabilities() {
+        // Base OpenAI-compatible profile. Concrete providers (Groq, Ollama, inference-snaps)
+        // override per their endpoint and default model.
+        return {
+            providerTag: 'openai-compat',
+            tools: true,
+            parallelToolCalls: false,
+            vision: false,
+            streaming: true,
+            embeddings: true,
+            reasoningEffort: false,
+            promptCache: false,
+            structuredOutput: false,
+        };
+    }
     async chat(messages, options) {
         const response = await fetch(`${this.baseURL}/chat/completions`, {
             method: 'POST',

package/dist/llm/token-counter.d.ts

@@ -19,6 +19,24 @@ export interface CostEstimate {
     model: string;
     direction: 'input' | 'output';
 }
+/**
+ * Per-1M-token pricing (USD). Single source of truth for every cost path in the package:
+ * `estimateCost` (input/output) and `calculateCacheCost` (cache-utils, cacheWrite/cacheRead).
+ * Cache rates follow Anthropic's model (write ~125% of input, read ~10%); non-caching
+ * providers carry 0 (local models are free; cache cost is not modelled for hosted
+ * non-Anthropic providers here).
+ */
+export interface ModelPricing {
+    /** USD per 1M input tokens. */
+    input: number;
+    /** USD per 1M output tokens. */
+    output: number;
+    /** USD per 1M tokens written to the prompt cache (0 where unsupported/unmodelled). */
+    cacheWrite: number;
+    /** USD per 1M tokens read from the prompt cache (0 where unsupported/unmodelled). */
+    cacheRead: number;
+}
+export declare const MODEL_PRICING: Record<string, ModelPricing>;
 /**
  * Estimate token count for a string.
  */

package/dist/llm/token-counter.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"token-counter.d.ts","sourceRoot":"","sources":["../../src/llm/token-counter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAEnD,MAAM,WAAW,gBAAgB;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,YAAY;IAC3B,gBAAgB,EAAE,MAAM,CAAC;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,OAAO,GAAG,QAAQ,CAAC;CAC/B;AAiCD;;GAEG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,gBAAgB,CAGxF;AAED;;;GAGG;AACH,wBAAgB,aAAa,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,gBAAgB,CAQjG;AAED;;;GAGG;AACH,wBAAgB,YAAY,CAC1B,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,MAAM,EACb,SAAS,EAAE,OAAO,GAAG,QAAQ,GAC5B,YAAY,CASd;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,OAAO,EAAE,EACnB,KAAK,EAAE,MAAM,GACZ;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,gBAAgB,EAAE,MAAM,CAAA;CAAE,CAI9C"}
+{"version":3,"file":"token-counter.d.ts","sourceRoot":"","sources":["../../src/llm/token-counter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAC;AAEnD,MAAM,WAAW,gBAAgB;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,YAAY;IAC3B,gBAAgB,EAAE,MAAM,CAAC;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,OAAO,GAAG,QAAQ,CAAC;CAC/B;AAED;;;;;;GAMG;AACH,MAAM,WAAW,YAAY;IAC3B,+BAA+B;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,gCAAgC;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,sFAAsF;IACtF,UAAU,EAAE,MAAM,CAAC;IACnB,qFAAqF;IACrF,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAmBtD,CAAC;AAeF;;GAEG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,gBAAgB,CAGxF;AAED;;;GAGG;AACH,wBAAgB,aAAa,CAAC,QAAQ,EAAE,OAAO,EAAE,EAAE,OAAO,CAAC,EAAE;IAAE,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,GAAG,gBAAgB,CAQjG;AAED;;;GAGG;AACH,wBAAgB,YAAY,CAC1B,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,MAAM,EACb,SAAS,EAAE,OAAO,GAAG,QAAQ,GAC5B,YAAY,CASd;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,QAAQ,EAAE,OAAO,EAAE,EACnB,KAAK,EAAE,MAAM,GACZ;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,gBAAgB,EAAE,MAAM,CAAA;CAAE,CAI9C"}

package/dist/llm/token-counter.js

@@ -8,22 +8,25 @@
  * Limitation: actual token counts differ by model tokenizer. This is
  * accurate enough for budget tracking and context window management.
  */
-// Cost per 1M tokens (USD)  -  input/output pricing
-const MODEL_PRICING = {
-    // Anthropic
-    'claude-opus-4-6': { input: 15.0, output: 75.0 },
-    'claude-sonnet-4-6': { input: 3.0, output: 15.0 },
-    'claude-haiku-4-5-20251001': { input: 0.25, output: 1.25 },
+export const MODEL_PRICING = {
+    // Anthropic (current)
+    'claude-opus-4-6': { input: 15.0, output: 75.0, cacheWrite: 18.75, cacheRead: 1.5 },
+    'claude-sonnet-4-6': { input: 3.0, output: 15.0, cacheWrite: 3.75, cacheRead: 0.3 },
+    'claude-haiku-4-5-20251001': { input: 0.25, output: 1.25, cacheWrite: 0.3125, cacheRead: 0.025 },
+    // Anthropic (legacy 2024  -  retained for cache-cost callers/examples)
+    'claude-3-5-sonnet-20241022': { input: 3.0, output: 15.0, cacheWrite: 3.75, cacheRead: 0.3 },
+    'claude-3-5-haiku-20241022': { input: 1.0, output: 5.0, cacheWrite: 1.25, cacheRead: 0.1 },
+    'claude-3-opus-20240229': { input: 15.0, output: 75.0, cacheWrite: 18.75, cacheRead: 1.5 },
     // OpenAI
-    'gpt-4o': { input: 5.0, output: 15.0 },
-    'gpt-4o-mini': { input: 0.15, output: 0.6 },
+    'gpt-4o': { input: 5.0, output: 15.0, cacheWrite: 0, cacheRead: 0 },
+    'gpt-4o-mini': { input: 0.15, output: 0.6, cacheWrite: 0, cacheRead: 0 },
     // Groq (Qwen  -  Apache 2.0)
-    'qwen/qwen3-32b': { input: 0.59, output: 0.79 },
+    'qwen/qwen3-32b': { input: 0.59, output: 0.79, cacheWrite: 0, cacheRead: 0 },
     // Ollama (self-hosted  -  no cost)
-    'gemma4:e2b': { input: 0, output: 0 },
-    'gemma4:e4b': { input: 0, output: 0 },
-    'gemma4:26b': { input: 0, output: 0 },
-    'nomic-embed-text': { input: 0, output: 0 },
+    'gemma4:e2b': { input: 0, output: 0, cacheWrite: 0, cacheRead: 0 },
+    'gemma4:e4b': { input: 0, output: 0, cacheWrite: 0, cacheRead: 0 },
+    'gemma4:26b': { input: 0, output: 0, cacheWrite: 0, cacheRead: 0 },
+    'nomic-embed-text': { input: 0, output: 0, cacheWrite: 0, cacheRead: 0 },
 };
 function charsPerToken(model) {
     const lower = model.toLowerCase();

package/dist/orchestration/runtime.d.ts

@@ -4,24 +4,18 @@
  * Executes agent tasks with tool execution, memory management, and error handling
  */
 import type { LLMClient } from '../llm/client.js';
+import type { ReasoningEffort } from '../llm/providers/base.js';
 import type { AgentSkillProvider } from '../skills/integration/agent-skill-provider.js';
 import type { ApprovalCallback } from '../tools/base.js';
 import type { MCPToolSource, McpClientLike } from '../tools/mcp-adapter.js';
 import type { Agent, AgentResult, Task } from './agent.js';
 /**
- * Controls how much of the model's token budget is allocated to internal reasoning.
- * Applies to Anthropic Claude only  -  ignored by other providers.
- *
- * | Level   | Budget (tokens) | Use case                          |
- * |---------|-----------------|-----------------------------------|
- * | off     | 0               | Disabled (default)                |
- * | minimal | 512             | Simple tasks, cost-sensitive      |
- * | low     | 2 048           | Moderate complexity               |
- * | medium  | 8 000           | Multi-step reasoning              |
- * | high    | 16 000          | Complex planning                  |
- * | xhigh   | 31 999          | Maximum depth (expensive)         |
+ * Reasoning-depth hint for a task. Alias of the neutral `ReasoningEffort` — the agnostic
+ * Reasoner port's control vocabulary. Mapping a level to a provider's native control
+ * (e.g. a thinking-token budget) is the adapter's job; a provider that advertises
+ * `reasoningEffort: false` treats it as a no-op.
  */
-export type ThinkingLevel = 'off' | 'minimal' | 'low' | 'medium' | 'high' | 'xhigh';
+export type ThinkingLevel = ReasoningEffort;
 export interface RuntimeConfig {
     maxIterations?: number;
     timeout?: number;

package/dist/orchestration/runtime.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"runtime.d.ts","sourceRoot":"","sources":["../../src/orchestration/runtime.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAIH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAGlD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,+CAA+C,CAAC;AACxF,OAAO,KAAK,EAAE,gBAAgB,EAAoB,MAAM,kBAAkB,CAAC;AAE3E,OAAO,KAAK,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,yBAAyB,CAAC;AAG5E,OAAO,KAAK,EAAE,KAAK,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAE3D;;;;;;;;;;;;GAYG;AACH,MAAM,MAAM,aAAa,GAAG,KAAK,GAAG,SAAS,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,CAAC;AAWpF,MAAM,WAAW,aAAa;IAC5B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;OAIG;IACH,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B;;;;;;;;;OASG;IACH,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B;;;;;;;;;;;;;;;;;;;OAmBG;IACH,UAAU,CAAC,EAAE,aAAa,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,aAAa,CAAA;KAAE,CAAC,CAAC;IACpE;;;;OAIG;IACH,aAAa,CAAC,EAAE,kBAAkB,CAAC;IACnC;;;;OAIG;IACH,SAAS,CAAC,EAAE,OAAO,gCAAgC,EAAE,SAAS,CAAC;IAC/D;;;;OAIG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC;;;;OAIG;IACH,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;CAClC;AAED,qBAAa,YAAY;IACvB,SAAS,CAAC,MAAM,EAAE,aAAa,CAAC;IAChC,OAAO,CAAC,SAAS,CAAc;IAC/B,OAAO,CAAC,cAAc,CAAgD;IACtE,OAAO,CAAC,cAAc,CAAS;gBAEnB,MAAM,GAAE,aAAkB;IA0BtC;;OAEG;IACG,WAAW,CAAC,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,CAAC;YAqBzE,OAAO;IAoPrB;;OAEG;IACH,WAAW,CAAC,IAAI,EAAE,IAAI,GAAG,IAAI;IAI7B;;OAEG;IACG,YAAY,CAAC,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IAc9E;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IA+B9B;;OAEG;IACH,SAAS,IAAI;QACX,cAAc,EAAE,OAAO,CAAC;QACxB,WAAW,EAAE,MAAM,CAAC;QACpB,cAAc,EAAE,MAAM,CAAC;KACxB;CAOF"}
+{"version":3,"file":"runtime.d.ts","sourceRoot":"","sources":["../../src/orchestration/runtime.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAIH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,KAAK,EAAW,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAEzE,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,+CAA+C,CAAC;AACxF,OAAO,KAAK,EAAE,gBAAgB,EAAoB,MAAM,kBAAkB,CAAC;AAE3E,OAAO,KAAK,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,yBAAyB,CAAC;AAG5E,OAAO,KAAK,EAAE,KAAK,EAAE,WAAW,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC;AAE3D;;;;;GAKG;AACH,MAAM,MAAM,aAAa,GAAG,eAAe,CAAC;AAE5C,MAAM,WAAW,aAAa;IAC5B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,6EAA6E;IAC7E,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB;;;;OAIG;IACH,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B;;;;;;;;;OASG;IACH,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B;;;;;;;;;;;;;;;;;;;OAmBG;IACH,UAAU,CAAC,EAAE,aAAa,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,aAAa,CAAA;KAAE,CAAC,CAAC;IACpE;;;;OAIG;IACH,aAAa,CAAC,EAAE,kBAAkB,CAAC;IACnC;;;;OAIG;IACH,SAAS,CAAC,EAAE,OAAO,gCAAgC,EAAE,SAAS,CAAC;IAC/D;;;;OAIG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf;;;OAGG;IACH,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC;;;;OAIG;IACH,qBAAqB,CAAC,EAAE,MAAM,EAAE,CAAC;CAClC;AAED,qBAAa,YAAY;IACvB,SAAS,CAAC,MAAM,EAAE,aAAa,CAAC;IAChC,OAAO,CAAC,SAAS,CAAc;IAC/B,OAAO,CAAC,cAAc,CAAgD;IACtE,OAAO,CAAC,cAAc,CAAS;gBAEnB,MAAM,GAAE,aAAkB;IA0BtC;;OAEG;IACG,WAAW,CAAC,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,CAAC;YAqBzE,OAAO;IA8OrB;;OAEG;IACH,WAAW,CAAC,IAAI,EAAE,IAAI,GAAG,IAAI;IAI7B;;OAEG;IACG,YAAY,CAAC,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IAc9E;;OAEG;IACG,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;IA+B9B;;OAEG;IACH,SAAS,IAAI;QACX,cAAc,EAAE,OAAO,CAAC;QACxB,WAAW,EAAE,MAAM,CAAC;QACpB,cAAc,EAAE,MAAM,CAAC;KACxB;CAOF"}

package/dist/orchestration/runtime.js

@@ -9,14 +9,6 @@ import { estimateCost } from '../llm/token-counter.js';
 import { ToolCallDeduplicator } from '../tools/deduplicator.js';
 import { createToolsFromMcpClient, discoverMCPTools } from '../tools/mcp-adapter.js';
 import { createWebSearchTool } from '../tools/web/duck-duck-go.js';
-const THINKING_BUDGETS = {
-    off: 0,
-    minimal: 512,
-    low: 2048,
-    medium: 8000,
-    high: 16000,
-    xhigh: 31999,
-};
 export class AgentRuntime {
     config;
     taskQueue = [];
@@ -101,8 +93,8 @@ export class AgentRuntime {
             {
                 role: 'system',
                 content: agent.instructions,
-                // Cache agent instructions for cost savings (Anthropic only)
-                cacheControl: this.config.enableCache ? { type: 'ephemeral' } : undefined,
+                // Cache agent instructions for cost savings (effective where promptCache is supported)
+                cache: this.config.enableCache || undefined,
             },
             {
                 role: 'user',
@@ -130,10 +122,6 @@ export class AgentRuntime {
                         },
                     };
                 }
-                // Resolve thinking budget if a level is configured
-                const thinkingBudget = this.config.thinkingLevel && this.config.thinkingLevel !== 'off'
-                    ? THINKING_BUDGETS[this.config.thinkingLevel]
-                    : undefined;
                 // Get LLM response (with caching for agent instructions and tools)
                 const response = await llmClient.chat(messages, {
                     tools: allTools.map((tool) => ({
@@ -144,8 +132,8 @@ export class AgentRuntime {
                             parameters: z.toJSONSchema(tool.parameters),
                         },
                     })),
-                    enableCache: this.config.enableCache,
-                    thinkingBudget,
+                    cacheHint: this.config.enableCache,
+                    effort: this.config.thinkingLevel,
                 });
                 // Accumulate token usage and cost
                 const iterationTokens = response.usage?.totalTokens ?? 0;

package/dist/orchestration/streaming-runtime.js

@@ -36,7 +36,7 @@ export class StreamingAgentRuntime extends AgentRuntime {
             {
                 role: 'system',
                 content: agent.instructions,
-                cacheControl: { type: 'ephemeral' },
+                cache: true,
             },
             {
                 role: 'user',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@revealui/ai",
-  "version": "0.6.2",
+  "version": "0.6.4",
   "description": "AI runtime for agent-driven products — agents, memory, LLM providers (Inference Snaps, Ollama, OpenAI-compatible), tools, and orchestration. Anthropic-SDK-free.",
   "keywords": [
     "agent",
@@ -25,20 +25,20 @@
   "dependencies": {
     "lru-cache": "^11.5.1",
     "zod": "^4.4.3",
-    "@revealui/contracts": "0.6.1",
-    "@revealui/core": "0.10.1",
-    "@revealui/db": "0.7.1",
+    "@revealui/contracts": "0.6.2",
+    "@revealui/core": "0.11.0",
+    "@revealui/db": "0.7.3",
     "@revealui/resilience": "0.2.4"
   },
   "devDependencies": {
-    "@electric-sql/pglite": "^0.5.1",
+    "@electric-sql/pglite": "^0.5.4",
     "@testing-library/react": "^16.3.2",
-    "@vitest/coverage-v8": "^4.1.8",
+    "@vitest/coverage-v8": "^4.1.10",
     "jsdom": "29.1.1",
     "react": "^19.2.7",
     "react-dom": "^19.2.7",
     "typescript": "^6.0.3",
-    "vitest": "^4.1.8",
+    "vitest": "^4.1.10",
     "@revealui/dev": "0.1.0"
   },
   "engines": {