@compilr-dev/sdk 0.18.9 → 0.18.10

package/dist/config.d.ts

@@ -49,7 +49,7 @@ export interface CapabilitiesConfig {
 /**
  * Supported provider types for auto-detection
  */
-export type ProviderType = 'claude' | 'openai' | 'gemini' | 'ollama' | 'together' | 'groq' | 'fireworks' | 'perplexity' | 'openrouter' | 'custom';
+export type ProviderType = 'claude' | 'openai' | 'gemini' | 'ollama' | 'ollama-anthropic' | 'together' | 'groq' | 'fireworks' | 'perplexity' | 'openrouter' | 'custom';
 /**
  * Tool configuration for controlling which tools are available
  */

package/dist/index.d.ts

@@ -45,7 +45,7 @@ export type { Preset } from './presets/index.js';
 export type { AnyTool } from './presets/types.js';
 export { resolveProvider, detectProviderFromEnv, createProviderFromType } from './provider.js';
 export { DEFAULT_MODELS, getContextWindow, DEFAULT_CONTEXT_WINDOW } from './models.js';
-export { type ModelTier, type TierInfo, type ProviderModelMap, MODEL_TIERS, TIER_INFO, isValidTier, type ThinkingFormat, type ModelStatus, type ModelInfo, MODEL_REGISTRY, getModelsForProvider, getModelsSortedForDisplay, getModelInfo, isKnownModel, isModelSupported, modelSupportsImages, modelSupportsTools, getThinkingFormat, getStatusIndicator, getStatusLabel, getDefaultModelForTier, areThinkingFormatsCompatible, shouldClearHistoryOnModelChange, getModelContextWindow, getModelDisplayName, getModelDescription, getModelForTier, getTierMappings, getTierDisplayName, getShortModelName, getDefaultTierMappings, type ProviderMetadata, type OthersProviderModel, PROVIDER_METADATA, TOGETHER_MODELS, GROQ_MODELS, FIREWORKS_MODELS, PERPLEXITY_MODELS, OPENROUTER_MODELS, getModelsForOthersProvider, getOthersProviders, getProviderMetadata, isOthersProvider, } from './models/index.js';
+export { type ModelTier, type TierInfo, type ProviderModelMap, MODEL_TIERS, TIER_INFO, isValidTier, type ThinkingFormat, type ModelStatus, type ModelInfo, MODEL_REGISTRY, getModelsForProvider, getModelsSortedForDisplay, getModelInfo, isKnownModel, isModelSupported, modelSupportsImages, modelSupportsTools, getThinkingFormat, getStatusIndicator, getStatusLabel, getDefaultModelForTier, areThinkingFormatsCompatible, shouldClearHistoryOnModelChange, getModelContextWindow, getModelDisplayName, getModelDescription, getModelForTier, getTierMappings, getTierDisplayName, getShortModelName, getDefaultTierMappings, type ProviderMetadata, type OthersProviderModel, PROVIDER_METADATA, TOGETHER_MODELS, GROQ_MODELS, FIREWORKS_MODELS, PERPLEXITY_MODELS, OPENROUTER_MODELS, getModelsForOthersProvider, getOthersProviders, getProviderMetadata, isOthersProvider, type OllamaModelInfo, type OllamaModelsResult, OLLAMA_ANTHROPIC_MIN_VERSION, listOllamaModels, getOllamaVersion, isOllamaAnthropicCompatible, clearOllamaModelsCache, formatOllamaModelSize, getOllamaModelDisplayName, getOllamaModelDescription, } from './models/index.js';
 export { assembleTools, deduplicateTools } from './tools.js';
 export { MetaToolsRegistry, createMetaTools, META_TOOLS_SYSTEM_PROMPT_PREFIX, } from './meta-tools/index.js';
 export type { MetaToolStats, MetaTools, FallbackOptions } from './meta-tools/index.js';

package/dist/index.js

@@ -85,7 +85,7 @@ export { DEFAULT_MODELS, getContextWindow, DEFAULT_CONTEXT_WINDOW } from './mode
 // =============================================================================
 export { MODEL_TIERS, TIER_INFO, isValidTier, MODEL_REGISTRY, getModelsForProvider, getModelsSortedForDisplay, getModelInfo, isKnownModel, isModelSupported, modelSupportsImages, modelSupportsTools, getThinkingFormat, getStatusIndicator, getStatusLabel, getDefaultModelForTier, areThinkingFormatsCompatible, shouldClearHistoryOnModelChange, getModelContextWindow, getModelDisplayName, getModelDescription, 
 // Model tiers (pure, settings-free)
-getModelForTier, getTierMappings, getTierDisplayName, getShortModelName, getDefaultTierMappings, PROVIDER_METADATA, TOGETHER_MODELS, GROQ_MODELS, FIREWORKS_MODELS, PERPLEXITY_MODELS, OPENROUTER_MODELS, getModelsForOthersProvider, getOthersProviders, getProviderMetadata, isOthersProvider, } from './models/index.js';
+getModelForTier, getTierMappings, getTierDisplayName, getShortModelName, getDefaultTierMappings, PROVIDER_METADATA, TOGETHER_MODELS, GROQ_MODELS, FIREWORKS_MODELS, PERPLEXITY_MODELS, OPENROUTER_MODELS, getModelsForOthersProvider, getOthersProviders, getProviderMetadata, isOthersProvider, OLLAMA_ANTHROPIC_MIN_VERSION, listOllamaModels, getOllamaVersion, isOllamaAnthropicCompatible, clearOllamaModelsCache, formatOllamaModelSize, getOllamaModelDisplayName, getOllamaModelDescription, } from './models/index.js';
 // =============================================================================
 // Tool Assembly
 // =============================================================================

package/dist/models/index.d.ts

@@ -5,3 +5,4 @@ export { type ModelTier, type TierInfo, type ProviderModelMap, MODEL_TIERS, TIER
 export { type ThinkingFormat, type ModelStatus, type ModelInfo, MODEL_REGISTRY, getModelsForProvider, getModelsSortedForDisplay, getModelInfo, isKnownModel, isModelSupported, modelSupportsImages, modelSupportsTools, getThinkingFormat, getStatusIndicator, getStatusLabel, getDefaultModelForTier, areThinkingFormatsCompatible, shouldClearHistoryOnModelChange, getModelContextWindow, getModelDisplayName, getModelDescription, } from './model-registry.js';
 export { getModelForTier, getTierMappings, getTierDisplayName, getShortModelName, getDefaultTierMappings, } from './model-tiers.js';
 export { type ProviderMetadata, type OthersProviderModel, PROVIDER_METADATA, TOGETHER_MODELS, GROQ_MODELS, FIREWORKS_MODELS, PERPLEXITY_MODELS, OPENROUTER_MODELS, getModelsForOthersProvider, getOthersProviders, getProviderMetadata, isOthersProvider, } from './providers.js';
+export { type OllamaModelInfo, type OllamaModelsResult, OLLAMA_ANTHROPIC_MIN_VERSION, listOllamaModels, getOllamaVersion, isOllamaAnthropicCompatible, clearOllamaModelsCache, formatOllamaModelSize, getOllamaModelDisplayName, getOllamaModelDescription, } from './ollama-discovery.js';

package/dist/models/index.js

@@ -9,3 +9,5 @@ export { MODEL_REGISTRY, getModelsForProvider, getModelsSortedForDisplay, getMod
 export { getModelForTier, getTierMappings, getTierDisplayName, getShortModelName, getDefaultTierMappings, } from './model-tiers.js';
 // Provider metadata
 export { PROVIDER_METADATA, TOGETHER_MODELS, GROQ_MODELS, FIREWORKS_MODELS, PERPLEXITY_MODELS, OPENROUTER_MODELS, getModelsForOthersProvider, getOthersProviders, getProviderMetadata, isOthersProvider, } from './providers.js';
+// Ollama discovery (host-agnostic — list installed models + version gate)
+export { OLLAMA_ANTHROPIC_MIN_VERSION, listOllamaModels, getOllamaVersion, isOllamaAnthropicCompatible, clearOllamaModelsCache, formatOllamaModelSize, getOllamaModelDisplayName, getOllamaModelDescription, } from './ollama-discovery.js';

package/dist/models/model-registry.js

@@ -684,6 +684,7 @@ function getProviderContextLimitFallback(provider) {
         case 'openai':
             return 128000;
         case 'ollama':
+        case 'ollama-anthropic':
             return 32000;
         case 'together':
             return 131072;

package/dist/models/ollama-discovery.d.ts

@@ -0,0 +1,63 @@
+/**
+ * Ollama discovery — host-agnostic helpers to list locally-pulled models and
+ * detect the Ollama version. Shared by CLI + Desktop so both surface installed
+ * models (closing the Desktop parity gap) for BOTH ollama providers
+ * (`ollama` OpenAI-compat and `ollama-anthropic`). Uses Ollama's native
+ * `/api/tags` and `/api/version` (same on both providers — same port).
+ *
+ * The base URL is a parameter (hosts pass their configured value); this module
+ * has no settings dependency. Results are cached per base URL for 30s.
+ * See ollama-anthropic-provider-spec.md §6.
+ */
+/** Minimum Ollama version that supports the native Anthropic `/v1/messages` endpoint. */
+export declare const OLLAMA_ANTHROPIC_MIN_VERSION = "0.14.0";
+/** Information about an Ollama model as returned by `/api/tags`. */
+export interface OllamaModelInfo {
+    /** Model name/tag (e.g., "llama3.3:70b", "ornith:latest") */
+    name: string;
+    /** Model size in bytes */
+    size: number;
+    /** When the model was modified/pulled */
+    modifiedAt: string;
+    /** Model digest (hash) */
+    digest: string;
+    /** Model details (family, parameter size, quantization) */
+    details?: {
+        family?: string;
+        parameterSize?: string;
+        quantizationLevel?: string;
+    };
+}
+/** Result of listing Ollama models. */
+export interface OllamaModelsResult {
+    success: boolean;
+    models: OllamaModelInfo[];
+    error?: string;
+    /** Whether the Ollama server responded at all */
+    ollamaRunning: boolean;
+    fetchedAt: number;
+}
+/** Clear the models cache (testing / forced refresh). */
+export declare function clearOllamaModelsCache(): void;
+/**
+ * List locally-available Ollama models via `/api/tags`. Cached per base URL for
+ * 30s. Returns gracefully (success:false) when Ollama isn't running.
+ */
+export declare function listOllamaModels(baseUrl?: string, forceRefresh?: boolean): Promise<OllamaModelsResult>;
+/**
+ * Get the running Ollama server version via `/api/version`, or null if
+ * unreachable. Used to gate the `ollama-anthropic` provider.
+ */
+export declare function getOllamaVersion(baseUrl?: string): Promise<string | null>;
+/**
+ * Whether the Ollama server supports the native Anthropic `/v1/messages`
+ * endpoint (version >= 0.14.0). Returns false if unreachable/unknown — callers
+ * should treat that as "offer classic ollama, gate/warn on ollama-anthropic".
+ */
+export declare function isOllamaAnthropicCompatible(baseUrl?: string): Promise<boolean>;
+/** Format a byte size for display (e.g. "4.1 GB"). */
+export declare function formatOllamaModelSize(bytes: number): string;
+/** Friendly display name for a model tag (e.g. "ornith:latest" → "Ornith (latest)"). */
+export declare function getOllamaModelDisplayName(name: string): string;
+/** Short description from model metadata (params · quant · size). */
+export declare function getOllamaModelDescription(model: OllamaModelInfo): string;

package/dist/models/ollama-discovery.js

@@ -0,0 +1,172 @@
+/**
+ * Ollama discovery — host-agnostic helpers to list locally-pulled models and
+ * detect the Ollama version. Shared by CLI + Desktop so both surface installed
+ * models (closing the Desktop parity gap) for BOTH ollama providers
+ * (`ollama` OpenAI-compat and `ollama-anthropic`). Uses Ollama's native
+ * `/api/tags` and `/api/version` (same on both providers — same port).
+ *
+ * The base URL is a parameter (hosts pass their configured value); this module
+ * has no settings dependency. Results are cached per base URL for 30s.
+ * See ollama-anthropic-provider-spec.md §6.
+ */
+const DEFAULT_BASE_URL = 'http://localhost:11434';
+/** Minimum Ollama version that supports the native Anthropic `/v1/messages` endpoint. */
+export const OLLAMA_ANTHROPIC_MIN_VERSION = '0.14.0';
+const CACHE_DURATION_MS = 30_000;
+const modelsCache = new Map();
+function resolveBaseUrl(baseUrl) {
+    return baseUrl ?? process.env.OLLAMA_BASE_URL ?? process.env.OLLAMA_HOST ?? DEFAULT_BASE_URL;
+}
+/** Clear the models cache (testing / forced refresh). */
+export function clearOllamaModelsCache() {
+    modelsCache.clear();
+}
+/**
+ * List locally-available Ollama models via `/api/tags`. Cached per base URL for
+ * 30s. Returns gracefully (success:false) when Ollama isn't running.
+ */
+export async function listOllamaModels(baseUrl, forceRefresh = false) {
+    const host = resolveBaseUrl(baseUrl);
+    const cached = modelsCache.get(host);
+    if (!forceRefresh && cached && Date.now() - cached.fetchedAt < CACHE_DURATION_MS) {
+        return cached;
+    }
+    try {
+        const response = await fetch(`${host}/api/tags`, {
+            method: 'GET',
+            signal: AbortSignal.timeout(5000),
+        });
+        if (!response.ok) {
+            const result = {
+                success: false,
+                models: [],
+                error: `Ollama returned status ${String(response.status)}`,
+                ollamaRunning: true,
+                fetchedAt: Date.now(),
+            };
+            modelsCache.set(host, result);
+            return result;
+        }
+        const data = (await response.json());
+        const models = (data.models ?? [])
+            .map((m) => ({
+            name: m.name,
+            size: m.size,
+            modifiedAt: m.modified_at,
+            digest: m.digest,
+            details: m.details
+                ? {
+                    family: m.details.family,
+                    parameterSize: m.details.parameter_size,
+                    quantizationLevel: m.details.quantization_level,
+                }
+                : undefined,
+        }))
+            .sort((a, b) => a.name.localeCompare(b.name));
+        const result = {
+            success: true,
+            models,
+            ollamaRunning: true,
+            fetchedAt: Date.now(),
+        };
+        modelsCache.set(host, result);
+        return result;
+    }
+    catch (error) {
+        let errorMsg = 'Unknown error';
+        const ollamaRunning = false;
+        if (error instanceof Error) {
+            if (error.name === 'AbortError' || error.message.includes('timeout')) {
+                errorMsg = 'Connection timed out';
+            }
+            else if (error.message.includes('ECONNREFUSED') || error.message.includes('fetch failed')) {
+                errorMsg = 'Ollama is not running. Start it with: ollama serve';
+            }
+            else {
+                errorMsg = error.message;
+            }
+        }
+        const result = {
+            success: false,
+            models: [],
+            error: errorMsg,
+            ollamaRunning,
+            fetchedAt: Date.now(),
+        };
+        modelsCache.set(host, result);
+        return result;
+    }
+}
+/**
+ * Get the running Ollama server version via `/api/version`, or null if
+ * unreachable. Used to gate the `ollama-anthropic` provider.
+ */
+export async function getOllamaVersion(baseUrl) {
+    const host = resolveBaseUrl(baseUrl);
+    try {
+        const response = await fetch(`${host}/api/version`, {
+            method: 'GET',
+            signal: AbortSignal.timeout(5000),
+        });
+        if (!response.ok)
+            return null;
+        const data = (await response.json());
+        return data.version ?? null;
+    }
+    catch {
+        return null;
+    }
+}
+/** Compare dotted numeric versions: -1 / 0 / 1. Non-numeric segments treated as 0. */
+function compareVersions(a, b) {
+    const pa = a.split('.').map((n) => parseInt(n, 10) || 0);
+    const pb = b.split('.').map((n) => parseInt(n, 10) || 0);
+    const len = Math.max(pa.length, pb.length);
+    for (let i = 0; i < len; i++) {
+        const d = (pa[i] ?? 0) - (pb[i] ?? 0);
+        if (d !== 0)
+            return d < 0 ? -1 : 1;
+    }
+    return 0;
+}
+/**
+ * Whether the Ollama server supports the native Anthropic `/v1/messages`
+ * endpoint (version >= 0.14.0). Returns false if unreachable/unknown — callers
+ * should treat that as "offer classic ollama, gate/warn on ollama-anthropic".
+ */
+export async function isOllamaAnthropicCompatible(baseUrl) {
+    const version = await getOllamaVersion(baseUrl);
+    if (!version)
+        return false;
+    return compareVersions(version, OLLAMA_ANTHROPIC_MIN_VERSION) >= 0;
+}
+// ── Display helpers (shared so hosts format installed models consistently) ──
+/** Format a byte size for display (e.g. "4.1 GB"). */
+export function formatOllamaModelSize(bytes) {
+    if (bytes < 1024)
+        return `${String(bytes)} B`;
+    if (bytes < 1024 * 1024)
+        return `${(bytes / 1024).toFixed(1)} KB`;
+    if (bytes < 1024 * 1024 * 1024)
+        return `${(bytes / (1024 * 1024)).toFixed(1)} MB`;
+    return `${(bytes / (1024 * 1024 * 1024)).toFixed(1)} GB`;
+}
+/** Friendly display name for a model tag (e.g. "ornith:latest" → "Ornith (latest)"). */
+export function getOllamaModelDisplayName(name) {
+    const [baseName, tag] = name.split(':');
+    const formattedBase = baseName
+        .split(/[-_]/)
+        .map((w) => w.charAt(0).toUpperCase() + w.slice(1))
+        .join(' ');
+    return tag ? `${formattedBase} (${tag})` : formattedBase;
+}
+/** Short description from model metadata (params · quant · size). */
+export function getOllamaModelDescription(model) {
+    const parts = [];
+    if (model.details?.parameterSize)
+        parts.push(model.details.parameterSize);
+    if (model.details?.quantizationLevel)
+        parts.push(model.details.quantizationLevel);
+    parts.push(formatOllamaModelSize(model.size));
+    return parts.join(' · ');
+}

package/dist/models/providers.js

@@ -56,6 +56,16 @@ export const PROVIDER_METADATA = {
         requiresKey: false,
         category: 'local',
     },
+    'ollama-anthropic': {
+        type: 'ollama-anthropic',
+        displayName: 'Ollama (Claude API)',
+        description: "Local models via Ollama's Anthropic endpoint (thinking; needs Ollama ≥ 0.14)",
+        endpoint: 'http://localhost:11434',
+        envVar: '',
+        keyUrl: 'https://ollama.ai',
+        requiresKey: false,
+        category: 'local',
+    },
     // "Others" providers (OpenAI-compatible APIs)
     together: {
         type: 'together',

package/dist/provider.js

@@ -48,6 +48,22 @@ export function createProviderFromType(type, options) {
             return createGeminiNativeProvider({ model, apiKey, estimateTokens, maxTokens });
         case 'ollama':
             return createOllamaProvider({ model, baseUrl, estimateTokens, maxTokens });
+        case 'ollama-anthropic':
+            // Ollama's native Anthropic Messages endpoint (Ollama >= 0.14). Reuses the
+            // Claude provider pointed at the Ollama base URL — the Anthropic SDK POSTs
+            // to {baseURL}/v1/messages. Anthropic-only features Ollama doesn't support
+            // are disabled; the dummy key is accepted-not-validated. Surfaces thinking
+            // that the OpenAI-compat `ollama` path drops. See ollama-anthropic spec.
+            return createClaudeProvider({
+                model,
+                apiKey: apiKey || 'ollama',
+                baseURL: baseUrl ?? 'http://localhost:11434',
+                estimateTokens,
+                maxTokens,
+                enablePromptCaching: false,
+                enableTokenEfficientTools: false,
+                enableExtendedContext: false,
+            });
         case 'together':
             return createTogetherProvider({ model, apiKey, estimateTokens, maxTokens });
         case 'groq':

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@compilr-dev/sdk",
-  "version": "0.18.9",
+  "version": "0.18.10",
   "description": "Universal agent runtime for building AI-powered applications",
   "type": "module",
   "main": "dist/index.js",