adaptive-memory-multi-model-router 1.9.5 → 2.0.0

package/src/providers/providerConfig.ts

@@ -0,0 +1,923 @@
+/**
+ * A3M Router - Generic Provider Configuration System
+ *
+ * Users can configure their available LLM providers via:
+ * 1. Environment variables (*_API_KEY patterns)
+ * 2. Config file at ~/.config/a3m-router/providers.json
+ * 3. Runtime registration via registerProvider()
+ *
+ * 40+ providers across free, cheap, mid-tier, premium, and enterprise tiers.
+ */
+
+import * as fs from 'fs';
+import * as path from 'path';
+
+// ============================================================
+// TYPES
+// ============================================================
+
+export type ProviderTier = 'free' | 'cheap' | 'mid' | 'premium' | 'enterprise';
+export type ProviderFormat = 'openai' | 'anthropic' | 'google' | 'cohere' | 'aws-bedrock' | 'google-vertex';
+export type ProviderType = 'api' | 'cli' | 'local';
+
+export interface ProviderCost {
+  input: number;  // per 1M tokens
+  output: number; // per 1M tokens
+}
+
+export interface ProviderDefinition {
+  id: string;
+  name: string;
+  baseUrl: string;
+  apiKeyEnv: string;
+  models: string[];
+  costPerK: ProviderCost;
+  tier: ProviderTier;
+  format: ProviderFormat;
+  type: ProviderType;
+  priority: number;
+  maxTokens: number;
+  cliCommand?: string;
+  apiKey?: string | null;
+}
+
+// ============================================================
+// DEFAULT PROVIDER DEFINITIONS
+// ============================================================
+
+export const DEFAULT_PROVIDERS: Record<string, ProviderDefinition> = {
+  // ========================================================================
+  // TIER: FREE / LOCAL
+  // ========================================================================
+  ollama: {
+    id: 'ollama',
+    name: 'Ollama',
+    baseUrl: 'http://127.0.0.1:11434/v1/chat/completions',
+    apiKeyEnv: '',
+    models: ['llama3', 'mistral', 'qwen2', 'codellama', 'phi3', 'gemma2'],
+    costPerK: { input: 0, output: 0 },
+    tier: 'free',
+    format: 'openai',
+    type: 'local',
+    priority: 1,
+    maxTokens: 8192,
+  },
+
+  lmstudio: {
+    id: 'lmstudio',
+    name: 'LM Studio',
+    baseUrl: 'http://127.0.0.1:1234/v1/chat/completions',
+    apiKeyEnv: '',
+    models: [],
+    costPerK: { input: 0, output: 0 },
+    tier: 'free',
+    format: 'openai',
+    type: 'local',
+    priority: 2,
+    maxTokens: 8192,
+  },
+
+  vllm: {
+    id: 'vllm',
+    name: 'vLLM',
+    baseUrl: 'http://127.0.0.1:8000/v1/chat/completions',
+    apiKeyEnv: '',
+    models: [],
+    costPerK: { input: 0, output: 0 },
+    tier: 'free',
+    format: 'openai',
+    type: 'local',
+    priority: 3,
+    maxTokens: 8192,
+  },
+
+  google: {
+    id: 'google',
+    name: 'Google AI',
+    baseUrl: 'https://generativelanguage.googleapis.com/v1beta/models',
+    apiKeyEnv: 'GOOGLE_API_KEY',
+    models: [
+      'gemini-2.5-flash',
+      'gemini-2.5-pro',
+      'gemini-2.0-flash',
+      'gemini-1.5-flash',
+      'gemini-1.5-pro',
+      'gemma-3-27b-it',
+    ],
+    costPerK: { input: 0, output: 0 }, // Free tier available
+    tier: 'free',
+    format: 'google',
+    type: 'api',
+    priority: 4,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // TIER: CHEAP / FAST (inference-optimized)
+  // ========================================================================
+  groq: {
+    id: 'groq',
+    name: 'Groq',
+    baseUrl: 'https://api.groq.com/openai/v1/chat/completions',
+    apiKeyEnv: 'GROQ_API_KEY',
+    models: [
+      'llama-3.3-70b-versatile',
+      'llama-3.1-8b-instant',
+      'openai/gpt-oss-120b',
+      'openai/gpt-oss-20b',
+      'qwen/qwen3-32b',
+      'meta-llama/llama-4-scout-17b-16e-instruct',
+    ],
+    costPerK: { input: 0.59, output: 0.79 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 5,
+    maxTokens: 8192,
+  },
+
+  cerebras: {
+    id: 'cerebras',
+    name: 'Cerebras',
+    baseUrl: 'https://api.cerebras.ai/v1/chat/completions',
+    apiKeyEnv: 'CEREBRAS_API_KEY',
+    models: [
+      'llama3.1-8b',
+      'llama-3.3-70b',
+      'qwen-3-235b-a22b-instruct-2507',
+      'gpt-oss-120b',
+      'zai-glm-4.7',
+    ],
+    costPerK: { input: 0.6, output: 0.6 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 6,
+    maxTokens: 8192,
+  },
+
+  deepinfra: {
+    id: 'deepinfra',
+    name: 'DeepInfra',
+    baseUrl: 'https://api.deepinfra.com/v1/openai/chat/completions',
+    apiKeyEnv: 'DEEPINFRA_API_KEY',
+    models: [
+      'meta-llama/Meta-Llama-3.1-8B-Instruct',
+      'meta-llama/Meta-Llama-3.1-70B-Instruct',
+      'mistralai/Mixtral-8x7B-Instruct-v0.1',
+      'Qwen/Qwen2.5-72B-Instruct',
+      'google/gemma-2-27b-it',
+    ],
+    costPerK: { input: 0.05, output: 0.05 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 7,
+    maxTokens: 8192,
+  },
+
+  together: {
+    id: 'together',
+    name: 'Together AI',
+    baseUrl: 'https://api.together.xyz/v1/chat/completions',
+    apiKeyEnv: 'TOGETHER_API_KEY',
+    models: [
+      'meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo',
+      'meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo',
+      'mistralai/Mixtral-8x7B-Instruct-v0.1',
+      'Qwen/Qwen2.5-72B-Instruct-Turbo',
+      'google/gemma-2-27b-it',
+    ],
+    costPerK: { input: 0.18, output: 0.18 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 8,
+    maxTokens: 8192,
+  },
+
+  fireworks: {
+    id: 'fireworks',
+    name: 'Fireworks AI',
+    baseUrl: 'https://api.fireworks.ai/inference/v1/chat/completions',
+    apiKeyEnv: 'FIREWORKS_API_KEY',
+    models: [
+      'accounts/fireworks/models/llama-v3p1-8b-instruct',
+      'accounts/fireworks/models/llama-v3p1-70b-instruct',
+      'accounts/fireworks/models/mixtral-8x7b-instruct',
+      'accounts/fireworks/models/qwen2p5-72b-instruct',
+    ],
+    costPerK: { input: 0.2, output: 0.2 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 9,
+    maxTokens: 8192,
+  },
+
+  novita: {
+    id: 'novita',
+    name: 'Novita AI',
+    baseUrl: 'https://api.novita.ai/v3/openai/chat/completions',
+    apiKeyEnv: 'NOVITA_API_KEY',
+    models: [
+      'meta-llama/llama-3.1-8b-instruct',
+      'meta-llama/llama-3.1-70b-instruct',
+      'Qwen/Qwen2.5-72B-Instruct',
+    ],
+    costPerK: { input: 0.06, output: 0.06 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 10,
+    maxTokens: 8192,
+  },
+
+  sambanova: {
+    id: 'sambanova',
+    name: 'SambaNova',
+    baseUrl: 'https://api.sambanova.ai/v1/chat/completions',
+    apiKeyEnv: 'SAMBANOVA_API_KEY',
+    models: [
+      'Meta-Llama-3.1-8B-Instruct',
+      'Meta-Llama-3.1-70B-Instruct',
+    ],
+    costPerK: { input: 0.1, output: 0.1 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 11,
+    maxTokens: 8192,
+  },
+
+  anyscale: {
+    id: 'anyscale',
+    name: 'Anyscale',
+    baseUrl: 'https://api.endpoints.anyscale.com/v1/chat/completions',
+    apiKeyEnv: 'ANYSCALE_API_KEY',
+    models: [
+      'meta-llama/Meta-Llama-3.1-8B-Instruct',
+      'meta-llama/Meta-Llama-3.1-70B-Instruct',
+      'mistralai/Mixtral-8x7B-Instruct-v0.1',
+    ],
+    costPerK: { input: 0.15, output: 0.15 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 12,
+    maxTokens: 8192,
+  },
+
+  replicate: {
+    id: 'replicate',
+    name: 'Replicate',
+    baseUrl: 'https://api.replicate.com/v1/chat/completions',
+    apiKeyEnv: 'REPLICATE_API_KEY',
+    models: [
+      'meta/llama-2-70b-chat',
+      'mistralai/mixtral-8x7b-instruct-v0.1',
+    ],
+    costPerK: { input: 0.2, output: 0.2 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 13,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // TIER: MID (good quality/price ratio)
+  // ========================================================================
+  deepseek: {
+    id: 'deepseek',
+    name: 'DeepSeek',
+    baseUrl: 'https://api.deepseek.com/v1/chat/completions',
+    apiKeyEnv: 'DEEPSEEK_API_KEY',
+    models: ['deepseek-chat', 'deepseek-reasoner', 'deepseek-coder'],
+    costPerK: { input: 0.14, output: 0.28 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 14,
+    maxTokens: 8192,
+  },
+
+  mistral: {
+    id: 'mistral',
+    name: 'Mistral',
+    baseUrl: 'https://api.mistral.ai/v1/chat/completions',
+    apiKeyEnv: 'MISTRAL_API_KEY',
+    models: [
+      'mistral-small-latest',
+      'mistral-medium-latest',
+      'mistral-large-latest',
+      'mistral-small-2506',
+      'devstral-small-2507',
+      'ministral-3b-latest',
+      'ministral-8b-latest',
+      'codestral-latest',
+      'open-mistral-nemo',
+    ],
+    costPerK: { input: 0.2, output: 0.6 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 15,
+    maxTokens: 8192,
+  },
+
+  perplexity: {
+    id: 'perplexity',
+    name: 'Perplexity',
+    baseUrl: 'https://api.perplexity.ai/chat/completions',
+    apiKeyEnv: 'PERPLEXITY_API_KEY',
+    models: [
+      'llama-3.1-sonar-small-128k-online',
+      'llama-3.1-sonar-large-128k-online',
+      'sonar-pro',
+    ],
+    costPerK: { input: 1.0, output: 1.0 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 16,
+    maxTokens: 8192,
+  },
+
+  cohere: {
+    id: 'cohere',
+    name: 'Cohere',
+    baseUrl: 'https://api.cohere.ai/v1/chat/completions',
+    apiKeyEnv: 'COHERE_API_KEY',
+    models: ['command-r-plus', 'command-r', 'command-a-03-2025'],
+    costPerK: { input: 2.5, output: 10 },
+    tier: 'mid',
+    format: 'cohere',
+    type: 'api',
+    priority: 17,
+    maxTokens: 8192,
+  },
+
+  ai21: {
+    id: 'ai21',
+    name: 'AI21 Labs',
+    baseUrl: 'https://api.ai21.com/studio/v1/chat/completions',
+    apiKeyEnv: 'AI21_API_KEY',
+    models: ['jamba-1.5-mini', 'jamba-1.5-large'],
+    costPerK: { input: 0.2, output: 0.4 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 18,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // TIER: PREMIUM (high-quality frontier models)
+  // ========================================================================
+  openai: {
+    id: 'openai',
+    name: 'OpenAI',
+    baseUrl: 'https://api.openai.com/v1/chat/completions',
+    apiKeyEnv: 'OPENAI_API_KEY',
+    models: [
+      'gpt-4o',
+      'gpt-4o-mini',
+      'gpt-4-turbo',
+      'gpt-4',
+      'gpt-3.5-turbo',
+      'o1-preview',
+      'o1-mini',
+    ],
+    costPerK: { input: 2.5, output: 10 },
+    tier: 'premium',
+    format: 'openai',
+    type: 'api',
+    priority: 19,
+    maxTokens: 8192,
+  },
+
+  anthropic: {
+    id: 'anthropic',
+    name: 'Anthropic',
+    baseUrl: 'https://api.anthropic.com/v1/messages',
+    apiKeyEnv: 'ANTHROPIC_API_KEY',
+    models: [
+      'claude-sonnet-4-20250514',
+      'claude-3.5-sonnet',
+      'claude-3-opus',
+      'claude-3-haiku',
+    ],
+    costPerK: { input: 3, output: 15 },
+    tier: 'premium',
+    format: 'anthropic',
+    type: 'api',
+    priority: 20,
+    maxTokens: 8192,
+  },
+
+  xai: {
+    id: 'xai',
+    name: 'xAI',
+    baseUrl: 'https://api.x.ai/v1/chat/completions',
+    apiKeyEnv: 'XAI_API_KEY',
+    models: ['grok-3', 'grok-3-mini', 'grok-2', 'grok-2-mini'],
+    costPerK: { input: 3.0, output: 15.0 },
+    tier: 'premium',
+    format: 'openai',
+    type: 'api',
+    priority: 21,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // TIER: ENTERPRISE (cloud-managed models)
+  // ========================================================================
+  azure_openai: {
+    id: 'azure_openai',
+    name: 'Azure OpenAI',
+    baseUrl: 'https://{resource}.openai.azure.com/openai/deployments/{deployment}/chat/completions',
+    apiKeyEnv: 'AZURE_OPENAI_API_KEY',
+    models: ['gpt-4', 'gpt-35-turbo', 'gpt-4o'],
+    costPerK: { input: 3.0, output: 12.0 },
+    tier: 'enterprise',
+    format: 'openai',
+    type: 'api',
+    priority: 22,
+    maxTokens: 8192,
+  },
+
+  bedrock: {
+    id: 'bedrock',
+    name: 'AWS Bedrock',
+    baseUrl: '',
+    apiKeyEnv: 'AWS_ACCESS_KEY_ID',
+    models: [
+      'anthropic.claude-3-sonnet',
+      'anthropic.claude-3-haiku',
+      'meta.llama3-1-8b',
+      'meta.llama3-1-70b',
+      'mistral.mixtral-8x7b',
+    ],
+    costPerK: { input: 3.0, output: 15.0 },
+    tier: 'enterprise',
+    format: 'aws-bedrock',
+    type: 'api',
+    priority: 23,
+    maxTokens: 8192,
+  },
+
+  vertex: {
+    id: 'vertex',
+    name: 'Google Vertex AI',
+    baseUrl: '',
+    apiKeyEnv: 'GOOGLE_APPLICATION_CREDENTIALS',
+    models: [
+      'gemini-1.5-flash',
+      'gemini-1.5-pro',
+      'claude-3-sonnet',
+      'claude-3-haiku',
+    ],
+    costPerK: { input: 1.25, output: 5.0 },
+    tier: 'enterprise',
+    format: 'google-vertex',
+    type: 'api',
+    priority: 24,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // ASIAN PROVIDERS
+  // ========================================================================
+  zhipu: {
+    id: 'zhipu',
+    name: 'Zhipu AI (GLM)',
+    baseUrl: 'https://open.bigmodel.cn/api/paas/v4/chat/completions',
+    apiKeyEnv: 'ZHIPU_API_KEY',
+    models: ['glm-4-flash', 'glm-4-plus', 'glm-4-air', 'glm-4-long'],
+    costPerK: { input: 0.1, output: 0.1 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 25,
+    maxTokens: 8192,
+  },
+
+  moonshot: {
+    id: 'moonshot',
+    name: 'Moonshot (Kimi)',
+    baseUrl: 'https://api.moonshot.cn/v1/chat/completions',
+    apiKeyEnv: 'MOONSHOT_API_KEY',
+    models: ['moonshot-v1-8k', 'moonshot-v1-32k', 'moonshot-v1-128k'],
+    costPerK: { input: 0.14, output: 0.14 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 26,
+    maxTokens: 8192,
+  },
+
+  qwen: {
+    id: 'qwen',
+    name: 'Alibaba Qwen (DashScope)',
+    baseUrl: 'https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions',
+    apiKeyEnv: 'DASHSCOPE_API_KEY',
+    models: ['qwen-turbo', 'qwen-plus', 'qwen-max', 'qwen-long'],
+    costPerK: { input: 0.4, output: 1.2 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 27,
+    maxTokens: 8192,
+  },
+
+  yi: {
+    id: 'yi',
+    name: 'Yi (01.AI)',
+    baseUrl: 'https://api.lingyiwanwu.com/v1/chat/completions',
+    apiKeyEnv: 'YI_API_KEY',
+    models: ['yi-lightning', 'yi-large', 'yi-medium', 'yi-spark'],
+    costPerK: { input: 0.2, output: 0.2 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 28,
+    maxTokens: 8192,
+  },
+
+  baichuan: {
+    id: 'baichuan',
+    name: 'Baichuan',
+    baseUrl: 'https://api.baichuan-ai.com/v1/chat/completions',
+    apiKeyEnv: 'BAICHUAN_API_KEY',
+    models: ['Baichuan4', 'Baichuan3-Turbo', 'Baichuan3-Turbo-128k'],
+    costPerK: { input: 0.2, output: 0.2 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 29,
+    maxTokens: 8192,
+  },
+
+  minimax: {
+    id: 'minimax',
+    name: 'MiniMax',
+    baseUrl: 'https://api.minimax.chat/v1/text/chatcompletion_v2',
+    apiKeyEnv: 'MINIMAX_API_KEY',
+    models: ['MiniMax-Text-01', 'abab6.5s-chat'],
+    costPerK: { input: 0.1, output: 0.1 },
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 30,
+    maxTokens: 8192,
+  },
+
+  stepfun: {
+    id: 'stepfun',
+    name: 'StepFun',
+    baseUrl: 'https://api.stepfun.com/v1/chat/completions',
+    apiKeyEnv: 'STEPFUN_API_KEY',
+    models: ['step-1-8k', 'step-1-32k', 'step-2-16k'],
+    costPerK: { input: 0.2, output: 0.2 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 31,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // EUROPEAN PROVIDERS
+  // ========================================================================
+  alephalpha: {
+    id: 'alephalpha',
+    name: 'Aleph Alpha',
+    baseUrl: 'https://api.aleph-alpha.com/v1/chat/completions',
+    apiKeyEnv: 'ALEPH_ALPHA_API_KEY',
+    models: ['luminous-base', 'luminous-extended', 'luminous-supreme'],
+    costPerK: { input: 2.0, output: 2.0 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 32,
+    maxTokens: 8192,
+  },
+
+  deepset: {
+    id: 'deepset',
+    name: 'Deepset',
+    baseUrl: 'https://api.deepset.ai/v1/chat/completions',
+    apiKeyEnv: 'DEEPSET_API_KEY',
+    models: ['gpt-4', 'claude-3-sonnet'],
+    costPerK: { input: 3.0, output: 12.0 },
+    tier: 'mid',
+    format: 'openai',
+    type: 'api',
+    priority: 33,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // OPEN ROUTER / AGGREGATORS
+  // ========================================================================
+  openrouter: {
+    id: 'openrouter',
+    name: 'OpenRouter',
+    baseUrl: 'https://openrouter.ai/api/v1/chat/completions',
+    apiKeyEnv: 'OPENROUTER_API_KEY',
+    models: [
+      'openai/gpt-4o',
+      'anthropic/claude-3.5-sonnet',
+      'google/gemini-pro-1.5',
+      'meta-llama/llama-3.1-70b-instruct',
+      'mistralai/mistral-large',
+    ],
+    costPerK: { input: 0, output: 0 }, // Passthrough pricing
+    tier: 'cheap',
+    format: 'openai',
+    type: 'api',
+    priority: 34,
+    maxTokens: 8192,
+  },
+
+  // ========================================================================
+  // CLI PROVIDERS (local tools)
+  // ========================================================================
+  opencode: {
+    id: 'opencode',
+    name: 'OpenCode',
+    baseUrl: '',
+    apiKeyEnv: '',
+    cliCommand: 'opencode',
+    models: [],
+    costPerK: { input: 0, output: 0 },
+    tier: 'free',
+    format: 'openai',
+    type: 'cli',
+    priority: 35,
+    maxTokens: 8192,
+  },
+
+  commandcode: {
+    id: 'commandcode',
+    name: 'CommandCode',
+    baseUrl: 'https://api.commandcode.ai/v1/chat/completions',
+    apiKeyEnv: 'COMMANDCODE_API_KEY',
+    models: ['taste-1'],
+    costPerK: { input: 0, output: 0 },
+    tier: 'free',
+    format: 'openai',
+    type: 'cli',
+    cliCommand: 'commandcode',
+    priority: 36,
+    maxTokens: 8192,
+  },
+};
+
+// ============================================================
+// RUNTIME STATE
+// ============================================================
+
+let _registeredProviders: Record<string, ProviderDefinition> = { ...DEFAULT_PROVIDERS };
+let _configLoaded = false;
+
+// ============================================================
+// CONFIGURATION LOADING
+// ============================================================
+
+export function loadConfig(configPath?: string): Record<string, ProviderDefinition> {
+  const paths: string[] = [];
+
+  if (configPath && fs.existsSync(configPath)) {
+    paths.push(configPath);
+  }
+
+  const userConfig = path.join(
+    process.env.HOME || process.env.USERPROFILE || '.',
+    '.config', 'a3m-router', 'providers.json'
+  );
+  if (fs.existsSync(userConfig)) {
+    paths.push(userConfig);
+  }
+
+  const projectConfig = path.join(process.cwd(), 'a3m-providers.json');
+  if (fs.existsSync(projectConfig)) {
+    paths.push(projectConfig);
+  }
+
+  for (const p of paths) {
+    try {
+      const raw = fs.readFileSync(p, 'utf-8');
+      const config = JSON.parse(raw);
+      if (config.providers) {
+        for (const [id, provider] of Object.entries(config.providers)) {
+          const prov = provider as Partial<ProviderDefinition>;
+          if (_registeredProviders[id]) {
+            _registeredProviders[id] = { ..._registeredProviders[id], ...prov };
+          } else {
+            _registeredProviders[id] = {
+              id,
+              type: 'api',
+              priority: 50,
+              maxTokens: 8192,
+              costPerK: { input: 0, output: 0 },
+              tier: 'mid',
+              format: 'openai',
+              models: [],
+              name: id,
+              baseUrl: '',
+              apiKeyEnv: '',
+              ...prov,
+            } as ProviderDefinition;
+          }
+        }
+      }
+      _configLoaded = true;
+      break;
+    } catch {
+      // Skip invalid config files
+    }
+  }
+
+  // Load API keys from environment
+  for (const provider of Object.values(_registeredProviders)) {
+    if (provider.apiKeyEnv) {
+      provider.apiKey = process.env[provider.apiKeyEnv] || null;
+    }
+  }
+
+  return _registeredProviders;
+}
+
+export function getAvailableProviders(): Record<string, ProviderDefinition> {
+  if (!_configLoaded) {
+    loadConfig();
+  }
+
+  const available: Record<string, ProviderDefinition> = {};
+
+  for (const [id, provider] of Object.entries(_registeredProviders)) {
+    if (provider.type === 'api') {
+      if (provider.apiKey) {
+        available[id] = provider;
+      }
+    } else {
+      available[id] = provider;
+    }
+  }
+
+  return Object.entries(available)
+    .sort(([, a], [, b]) => a.priority - b.priority)
+    .reduce((acc: Record<string, ProviderDefinition>, [k, v]) => { acc[k] = v; return acc; }, {});
+}
+
+// ============================================================
+// RUNTIME REGISTRATION
+// ============================================================
+
+export function registerProvider(id: string, config: Partial<ProviderDefinition>): ProviderDefinition {
+  _registeredProviders[id] = {
+    id,
+    name: config.name || id,
+    baseUrl: config.baseUrl || '',
+    apiKeyEnv: config.apiKeyEnv || '',
+    type: 'api',
+    priority: 50,
+    maxTokens: 8192,
+    costPerK: { input: 0, output: 0 },
+    tier: 'mid',
+    format: 'openai',
+    models: [],
+    ...config,
+  } as ProviderDefinition;
+  return _registeredProviders[id];
+}
+
+export function deregisterProvider(id: string): void {
+  delete _registeredProviders[id];
+}
+
+export function updateProvider(id: string, updates: Partial<ProviderDefinition>): ProviderDefinition | null {
+  if (_registeredProviders[id]) {
+    _registeredProviders[id] = { ..._registeredProviders[id], ...updates };
+    return _registeredProviders[id];
+  }
+  return null;
+}
+
+// ============================================================
+// HEALTH CHECK
+// ============================================================
+
+export async function healthCheck(providerId: string): Promise<{
+  healthy: boolean;
+  error?: string;
+  latency?: number;
+  model?: string;
+  type?: string;
+}> {
+  const provider = _registeredProviders[providerId];
+  if (!provider) {
+    return { healthy: false, error: 'Provider not found: ' + providerId };
+  }
+
+  if (provider.type === 'cli') {
+    const { execSync } = require('child_process');
+    try {
+      execSync(`which ${provider.cliCommand || provider.id}`, { stdio: 'pipe' });
+      return { healthy: true, latency: 0, type: 'cli' };
+    } catch {
+      return { healthy: false, error: 'Command not found: ' + (provider.cliCommand || provider.id) };
+    }
+  }
+
+  if (provider.type === 'api') {
+    if (!provider.apiKey) {
+      return { healthy: false, error: 'No API key for ' + provider.name };
+    }
+
+    const startTime = Date.now();
+    try {
+      const model = provider.models[0];
+      const resp = await fetch(provider.baseUrl, {
+        method: 'POST',
+        headers: {
+          'Authorization': 'Bearer ' + provider.apiKey,
+          'Content-Type': 'application/json',
+        },
+        body: JSON.stringify({
+          model,
+          messages: [{ role: 'user', content: 'test' }],
+          max_tokens: 5,
+        }),
+      });
+
+      const latency = Date.now() - startTime;
+      const data = await resp.json() as any;
+
+      if (data.error) {
+        return { healthy: false, error: data.error.message, latency };
+      }
+
+      return { healthy: true, latency, model: data.model || model };
+    } catch (e: any) {
+      return { healthy: false, error: e.message, latency: Date.now() - startTime };
+    }
+  }
+
+  return { healthy: false, error: 'Unknown provider type: ' + provider.type };
+}
+
+export async function checkAllProviders(): Promise<Record<string, any>> {
+  const results: Record<string, any> = {};
+  const available = getAvailableProviders();
+
+  for (const id of Object.keys(available)) {
+    results[id] = await healthCheck(id);
+  }
+
+  return results;
+}
+
+export function findCheapestAvailableProvider(model?: string): ProviderDefinition | null {
+  const available = getAvailableProviders();
+  const sorted = Object.values(available).sort(
+    (a, b) => (a.costPerK.input + a.costPerK.output) - (b.costPerK.input + b.costPerK.output)
+  );
+  if (model) {
+    return sorted.find(p => p.models.includes(model)) || null;
+  }
+  return sorted[0] || null;
+}
+
+export function findFastestAvailableProvider(): ProviderDefinition | null {
+  const available = getAvailableProviders();
+  const cheapTier = Object.values(available).filter(p => p.tier === 'cheap');
+  return cheapTier[0] || Object.values(available)[0] || null;
+}
+
+// ============================================================
+// SAVE CONFIG
+// ============================================================
+
+export function saveConfig(configPath?: string): string {
+  const target = configPath || path.join(
+    process.env.HOME || '.',
+    '.config', 'a3m-router', 'providers.json'
+  );
+
+  const dir = path.dirname(target);
+  if (!fs.existsSync(dir)) {
+    fs.mkdirSync(dir, { recursive: true });
+  }
+
+  const safeConfig: Record<string, any> = {};
+  for (const [id, provider] of Object.entries(_registeredProviders)) {
+    safeConfig[id] = { ...provider };
+    delete safeConfig[id].apiKey;
+  }
+
+  fs.writeFileSync(target, JSON.stringify({ providers: safeConfig }, null, 2));
+  return target;
+}