pediatric-ai-scribe-v3/src/utils/models.js

// ============================================================
// MODELS.JS — Provider-aware model list
// Supports: OpenRouter, AWS Bedrock, Azure OpenAI, LiteLLM.
// Vertex and anything else Google is reached through LiteLLM, not directly.
// ============================================================

var activeProvider = process.env.AI_PROVIDER || (process.env.LITELLM_API_BASE ? 'litellm' : 'openrouter');

var OPENROUTER_MODELS = [
  { id: 'google/gemini-2.5-flash', name: 'Gemini Flash 2.5' },
  { id: 'google/gemini-2.5-pro', name: 'Gemini Pro 2.5' },
  { id: 'google/gemini-2.5-flash:thinking', name: 'Gemini Flash Thinking' },
  { id: 'deepseek/deepseek-chat-v3-0324', name: 'DeepSeek V3' },
  { id: 'deepseek/deepseek-r1', name: 'DeepSeek R1' },
  { id: 'deepseek/deepseek-r1:free', name: 'DeepSeek R1 Free' },
  { id: 'qwen/qwen3-235b-a22b', name: 'Qwen3 235B' },
  { id: 'qwen/qwen3-30b-a3b:free', name: 'Qwen3 30B Free' },
  { id: 'meta-llama/llama-4-maverick', name: 'Llama 4 Maverick' },
  { id: 'meta-llama/llama-3.3-70b-instruct', name: 'Llama 3.3 70B' },
  { id: 'openai/gpt-4.1', name: 'GPT-4.1' },
  { id: 'openai/gpt-4.1-mini', name: 'GPT-4.1 Mini' },
  { id: 'openai/o4-mini', name: 'o4-mini (Reasoning)' },
  { id: 'mistralai/mistral-large-2411', name: 'Mistral Large' },
  { id: 'mistralai/mistral-small-3.2-24b-instruct:free', name: 'Mistral Small Free' }
];

var BEDROCK_MODELS = [
  // All IDs verified against AWS docs 2026-03-24
  // us. prefix = cross-region inference profile (required for most newer models)
  // maxOut = model's max output token limit (omit if >= 8192)

  // ── Amazon Nova (inference profiles) ──
  { id: 'amazon/nova-pro', name: 'Amazon Nova Pro',
    bedrockId: 'us.amazon.nova-pro-v1:0' },
  { id: 'amazon/nova-lite', name: 'Amazon Nova Lite',
    bedrockId: 'us.amazon.nova-lite-v1:0' },
  { id: 'amazon/nova-micro', name: 'Amazon Nova Micro',
    bedrockId: 'us.amazon.nova-micro-v1:0' },

  // ── Meta Llama (inference profiles) ──
  { id: 'meta/llama-4-maverick', name: 'Llama 4 Maverick',
    bedrockId: 'us.meta.llama4-maverick-17b-instruct-v1:0' },
  { id: 'meta/llama-4-scout', name: 'Llama 4 Scout',
    bedrockId: 'us.meta.llama4-scout-17b-instruct-v1:0' },
  { id: 'meta/llama-3.3-70b', name: 'Llama 3.3 70B',
    bedrockId: 'us.meta.llama3-3-70b-instruct-v1:0' },

  // ── DeepSeek (R1 has profile, V3.2 is on-demand) ──
  { id: 'deepseek/r1', name: 'DeepSeek R1',
    bedrockId: 'us.deepseek.r1-v1:0' },
  { id: 'deepseek/v3.2', name: 'DeepSeek V3.2',
    bedrockId: 'deepseek.v3.2', regions: ['us-east-1', 'us-east-2', 'us-west-2'] },

  // ── Mistral AI (on-demand, no profiles available) ──
  { id: 'mistral/large-3', name: 'Mistral Large 3 (675B)',
    bedrockId: 'mistral.mistral-large-3-675b-instruct', regions: ['us-east-1', 'us-east-2', 'us-west-2'] },
  { id: 'mistral/magistral-small', name: 'Magistral Small',
    bedrockId: 'mistral.magistral-small-2509', regions: ['us-east-1', 'us-east-2', 'us-west-2'] },

  // ── Cohere (on-demand, max 4096 output) ──
  { id: 'cohere/command-r-plus', name: 'Command R+',
    bedrockId: 'cohere.command-r-plus-v1:0', regions: ['us-east-1', 'us-west-2'], maxOut: 4096 },
  { id: 'cohere/command-r', name: 'Command R',
    bedrockId: 'cohere.command-r-v1:0', regions: ['us-east-1', 'us-west-2'], maxOut: 4096 },

  // ── AI21 Labs (on-demand, max 4096 output) ──
  { id: 'ai21/jamba-1.5-large', name: 'Jamba 1.5 Large',
    bedrockId: 'ai21.jamba-1-5-large-v1:0', regions: ['us-east-1'], maxOut: 4096 },

  // ── Writer (inference profile) ──
  { id: 'writer/palmyra-x5', name: 'Palmyra X5',
    bedrockId: 'us.writer.palmyra-x5-v1:0' },

  // ── Qwen (on-demand, no profiles) ──
  { id: 'qwen/qwen3-235b', name: 'Qwen3 235B',
    bedrockId: 'qwen.qwen3-235b-a22b-2507-v1:0' },
  { id: 'qwen/qwen3-32b', name: 'Qwen3 32B',
    bedrockId: 'qwen.qwen3-32b-v1:0' }
];

var AZURE_MODELS = [
  { id: 'gpt-4o', name: 'GPT-4o',
    deploymentNote: 'Set AZURE_DEPLOYMENT_NAME in .env' },
  { id: 'gpt-4o-mini', name: 'GPT-4o Mini',
    deploymentNote: 'Set AZURE_DEPLOYMENT_NAME in .env' },
  { id: 'gpt-4.1', name: 'GPT-4.1',
    deploymentNote: 'Set AZURE_DEPLOYMENT_NAME in .env' },
  { id: 'gpt-4.1-mini', name: 'GPT-4.1 Mini',
    deploymentNote: 'Set AZURE_DEPLOYMENT_NAME in .env' }
];

// LiteLLM has NO built-in models — everything is discovered from the proxy via admin panel.
// This array is intentionally empty. Do not add models here.
var LITELLM_MODELS = [];

function getAvailableModels() {
  switch (activeProvider) {
    case 'bedrock':
      var region = process.env.AWS_BEDROCK_REGION || 'us-east-1';
      return BEDROCK_MODELS.filter(function(m) {
        return !m.regions || m.regions.indexOf(region) !== -1;
      });
    case 'azure': return AZURE_MODELS;
    case 'litellm': return LITELLM_MODELS;
    case 'openrouter':
    default: return OPENROUTER_MODELS;
  }
}

function getDefaultModel() {
  switch (activeProvider) {
    case 'bedrock': return 'amazon/nova-pro';
    case 'azure': return process.env.AZURE_DEPLOYMENT_NAME || 'gpt-4o-mini';
    case 'litellm': return process.env.LITELLM_DEFAULT_MODEL || process.env.OPENAI_MODEL || ''; // OpenAI-compatible base: set admin default or env.
    case 'openrouter':
    default: return 'google/gemini-2.5-flash';
  }
}

function getFallbackModel() {
  switch (activeProvider) {
    case 'bedrock': return 'amazon/nova-lite';
    case 'azure': return process.env.AZURE_DEPLOYMENT_NAME || 'gpt-4o-mini';
    case 'litellm': return process.env.LITELLM_FALLBACK_MODEL || ''; // No implicit cross-provider fallback.
    case 'openrouter':
    default: return 'deepseek/deepseek-chat-v3-0324';
  }
}

function getBedrockModelId(modelId) {
  var found = BEDROCK_MODELS.find(function(m) { return m.id === modelId; });
  return found ? found.bedrockId : modelId;
}

function getBedrockMaxOut(modelId) {
  var found = BEDROCK_MODELS.find(function(m) { return m.id === modelId; });
  return found && found.maxOut ? found.maxOut : null;
}


var AVAILABLE_MODELS = getAvailableModels();
var DEFAULT_MODEL = getDefaultModel();
var FALLBACK_MODEL = getFallbackModel();

// node:test reads a test file's results back over the child's stdout, so a
// banner written there can land inside a serialized frame and fail the whole
// file. NODE_TEST_CONTEXT is set only inside those children, so production and
// `node server.js` output is unchanged.
if (!process.env.NODE_TEST_CONTEXT) {
  console.log('🤖 Provider:', activeProvider);
  console.log('🤖 Default model:', DEFAULT_MODEL);
  console.log('🤖 Models available:', AVAILABLE_MODELS.length);
}

// DB-aware model list (used by /api/models endpoint)
async function getAvailableModelsWithOverrides(db) {
  var baseModels = getAvailableModels();
  var disabledRaw = await db.getSetting('models.disabled');
  var customRaw = await db.getSetting('models.custom');
  var disabled = JSON.parse(disabledRaw == null ? '[]' : disabledRaw);
  var custom = JSON.parse(customRaw == null ? '[]' : customRaw);
  if (!Array.isArray(disabled) || disabled.some(function(id) { return typeof id !== 'string' || !id.trim(); }) ||
      !Array.isArray(custom) || custom.some(function(m) { return !m || typeof m.id !== 'string' || !m.id.trim() || m.id !== m.id.trim(); })) {
    throw new Error('Invalid model settings');
  }
  var result = baseModels.slice();
  custom.forEach(function(m) {
    if (!result.some(function(r) { return r.id === m.id; })) result.push(m);
  });
  return result.filter(function(m) { return !disabled.includes(m.id); });
}

// Read policy on every call: no stale grants after edits or during a DB outage.
async function getAllowedModelIds(db) {
  var models = await getAvailableModelsWithOverrides(db);
  return new Set(models.map(function(m) { return m.id; }));
}

// Never advertise or use a stale/disabled default, including source/env defaults.
async function getEffectiveDefaultModel(db, models) {
  models = models || await getAvailableModelsWithOverrides(db);
  var saved = await db.getSetting('models.default');
  var preferred = saved || getDefaultModel();
  return models.some(function(m) { return m.id === preferred; }) ? preferred : (models[0] ? models[0].id : '');
}

async function reconcileDefaultModel(db) {
  var saved = await db.getSetting('models.default');
  if (saved && !(await getAvailableModelsWithOverrides(db)).some(function(m) { return m.id === saved; })) {
    await db.setSetting('models.default', '');
  }
}

// Synchronous check against the static baseline (no DB access). Used when
// the async-aware helper is inconvenient. For full enforcement use
// getAllowedModelIds + Set.has() in the route before invoking callAI.
function isStaticAllowedModel(id) {
  if (!id) return false;
  return getAvailableModels().some(function(m) { return m.id === id; });
}

module.exports = {
  AVAILABLE_MODELS,
  DEFAULT_MODEL,
  FALLBACK_MODEL,
  OPENROUTER_MODELS,
  BEDROCK_MODELS,
  AZURE_MODELS,
  LITELLM_MODELS,
  getAvailableModels,
  getAvailableModelsWithOverrides,
  getAllowedModelIds,
  getEffectiveDefaultModel,
  reconcileDefaultModel,
  isStaticAllowedModel,
  getDefaultModel,
  getFallbackModel,
  getBedrockModelId,
  getBedrockMaxOut,
  activeProvider
};