William Valentin
499 lines
17 KiB
TypeScript
499 lines
17 KiB
TypeScript
import type { Config, ModelConfig } from '../config/index.js';
|
|
import { AnthropicClient, OpenAIClient, OllamaClient, LlamaCppClient, GeminiClient, BedrockClient, GitHubModelsClient, SyntheticClient, RotatingModelClient, ModelRouter, DEFAULT_RETRY_CONFIG } from '../models/index.js';
|
|
import type { ModelClient, RetryConfig, ModelTier } from '../models/index.js';
|
|
import { logger } from '../logger.js';
|
|
import { getZaiApiKey } from '../auth/zai.js';
|
|
import { getAnthropicApiKey, getAnthropicAuthToken } from '../auth/anthropic.js';
|
|
import { getGeminiApiKey } from '../auth/gemini.js';
|
|
import { getOpenAIApiKey, loadStoredOpenAIAuth } from '../auth/openai.js';
|
|
|
|
type AuthMode = 'auto' | 'api_key' | 'oauth';
|
|
|
|
function getEffectiveAuthMode(cfg: ModelConfig): AuthMode {
|
|
if (cfg.auth_mode) {
|
|
return cfg.auth_mode;
|
|
}
|
|
if (cfg.use_oauth) {
|
|
return 'oauth';
|
|
}
|
|
return 'auto';
|
|
}
|
|
|
|
/**
|
|
* Resolve an API key from config or environment variable.
|
|
* Throws a clear error naming the expected env var if neither source provides a key.
|
|
*/
|
|
function requireApiKey(cfg: ModelConfig, envVar: string): string {
|
|
const key = cfg.api_key ?? process.env[envVar];
|
|
if (!key) {
|
|
throw new Error(
|
|
`API key required for ${cfg.provider}. ` +
|
|
`Set ${envVar} environment variable or provide api_key in config.`,
|
|
);
|
|
}
|
|
return key;
|
|
}
|
|
|
|
function resolveApiKeyPool(cfg: ModelConfig, envVar?: string): string[] {
|
|
const configured = (cfg.api_keys ?? []).map((key) => key.trim()).filter(Boolean);
|
|
if (configured.length > 0) {
|
|
return configured;
|
|
}
|
|
if (cfg.api_key?.trim()) {
|
|
return [cfg.api_key.trim()];
|
|
}
|
|
if (envVar && process.env[envVar]?.trim()) {
|
|
return [process.env[envVar]!.trim()];
|
|
}
|
|
return [];
|
|
}
|
|
|
|
function createApiKeyClient(
|
|
keys: string[],
|
|
build: (apiKey: string) => ModelClient,
|
|
options?: { cooldownMs?: number },
|
|
): ModelClient {
|
|
if (keys.length === 1) {
|
|
return build(keys[0]);
|
|
}
|
|
return new RotatingModelClient(
|
|
keys.map((key) => build(key)),
|
|
{ cooldownMs: options?.cooldownMs ?? 0 },
|
|
);
|
|
}
|
|
|
|
function resolveZaiCredential(cfg: ModelConfig): string {
|
|
const raw = cfg.api_key
|
|
?? cfg.auth_token
|
|
?? getZaiApiKey();
|
|
|
|
if (!raw) {
|
|
throw new Error(
|
|
'Z.AI credential not configured. ' +
|
|
'Run `flynn zai-auth` or set ZAI_API_KEY / ZHIPUAI_API_KEY / ZHIPUAI_AUTH_TOKEN, ' +
|
|
'or provide api_key/auth_token in config.',
|
|
);
|
|
}
|
|
|
|
return raw.startsWith('Bearer ') ? raw.slice('Bearer '.length) : raw;
|
|
}
|
|
|
|
/**
|
|
* Create a ModelClient from a provider config entry.
|
|
* Dispatches on the `provider` field so all tiers and fallback entries
|
|
* use the correct client implementation.
|
|
*/
|
|
export function createClientFromConfig(cfg: ModelConfig): ModelClient {
|
|
switch (cfg.provider) {
|
|
case 'anthropic':
|
|
{
|
|
const authMode = getEffectiveAuthMode(cfg);
|
|
|
|
if (authMode === 'oauth') {
|
|
const token = cfg.auth_token ?? getAnthropicAuthToken();
|
|
if (!token) {
|
|
throw new Error(
|
|
'Anthropic auth token not configured (auth_mode: oauth). ' +
|
|
'Set ANTHROPIC_AUTH_TOKEN, run `flynn anthropic-auth --token`, or provide auth_token in config.',
|
|
);
|
|
}
|
|
return new AnthropicClient({
|
|
model: cfg.model,
|
|
authToken: token,
|
|
});
|
|
}
|
|
|
|
if (authMode === 'api_key') {
|
|
const keys = resolveApiKeyPool(cfg);
|
|
const envKey = getAnthropicApiKey();
|
|
const allKeys = keys.length > 0
|
|
? keys
|
|
: (envKey ? [envKey] : []);
|
|
if (allKeys.length === 0) {
|
|
throw new Error(
|
|
'Anthropic API key not configured (auth_mode: api_key). ' +
|
|
'Set ANTHROPIC_API_KEY, run `flynn anthropic-auth`, or provide api_key/api_keys in config.',
|
|
);
|
|
}
|
|
return createApiKeyClient(allKeys, (apiKey) => new AnthropicClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
|
|
// auto: prefer API keys, then token
|
|
const configuredKeys = resolveApiKeyPool(cfg);
|
|
const envKey = getAnthropicApiKey();
|
|
const allKeys = configuredKeys.length > 0
|
|
? configuredKeys
|
|
: (envKey ? [envKey] : []);
|
|
if (allKeys.length > 0) {
|
|
return createApiKeyClient(allKeys, (apiKey) => new AnthropicClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
|
|
const token = cfg.auth_token ?? getAnthropicAuthToken();
|
|
if (token) {
|
|
return new AnthropicClient({
|
|
model: cfg.model,
|
|
authToken: token,
|
|
});
|
|
}
|
|
|
|
throw new Error(
|
|
'Anthropic credentials not configured (auth_mode: auto). ' +
|
|
'Set ANTHROPIC_API_KEY (or run `flynn anthropic-auth`), ' +
|
|
'or set ANTHROPIC_AUTH_TOKEN (or run `flynn anthropic-auth --token`).',
|
|
);
|
|
}
|
|
case 'openai':
|
|
{
|
|
const authMode = getEffectiveAuthMode(cfg);
|
|
|
|
if (authMode === 'oauth') {
|
|
const existing = loadStoredOpenAIAuth();
|
|
if (!existing) {
|
|
throw new Error(
|
|
'OpenAI OAuth is not configured (auth_mode: oauth). ' +
|
|
'Run `flynn openai-auth` to authenticate.',
|
|
);
|
|
}
|
|
return new OpenAIClient({
|
|
model: cfg.model,
|
|
useOAuth: true,
|
|
});
|
|
}
|
|
|
|
if (authMode === 'api_key') {
|
|
const keys = resolveApiKeyPool(cfg);
|
|
const envKey = getOpenAIApiKey();
|
|
const allKeys = keys.length > 0
|
|
? keys
|
|
: (envKey ? [envKey] : []);
|
|
if (allKeys.length === 0) {
|
|
throw new Error(
|
|
'OpenAI API key not configured (auth_mode: api_key). ' +
|
|
'Set OPENAI_API_KEY, run `flynn openai-key`, or provide api_key/api_keys in config.',
|
|
);
|
|
}
|
|
return createApiKeyClient(allKeys, (apiKey) => new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
|
|
// auto: prefer API keys, then OAuth
|
|
const configuredKeys = resolveApiKeyPool(cfg);
|
|
const envKey = getOpenAIApiKey();
|
|
const allKeys = configuredKeys.length > 0
|
|
? configuredKeys
|
|
: (envKey ? [envKey] : []);
|
|
if (allKeys.length > 0) {
|
|
return createApiKeyClient(allKeys, (apiKey) => new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
|
|
const existing = loadStoredOpenAIAuth();
|
|
if (existing) {
|
|
return new OpenAIClient({
|
|
model: cfg.model,
|
|
useOAuth: true,
|
|
});
|
|
}
|
|
|
|
throw new Error(
|
|
'OpenAI credentials not configured (auth_mode: auto). ' +
|
|
'Set OPENAI_API_KEY (or run `flynn openai-key`), ' +
|
|
'or run `flynn openai-auth` for OAuth.',
|
|
);
|
|
}
|
|
case 'ollama':
|
|
return new OllamaClient({
|
|
model: cfg.model,
|
|
host: cfg.endpoint,
|
|
numGpu: cfg.num_gpu,
|
|
});
|
|
case 'llamacpp':
|
|
return new LlamaCppClient({
|
|
endpoint: cfg.endpoint ?? 'http://localhost:8080',
|
|
model: cfg.model,
|
|
authToken: cfg.auth_token,
|
|
});
|
|
case 'gemini':
|
|
{
|
|
const apiKey = cfg.api_key ?? getGeminiApiKey() ?? undefined;
|
|
return new GeminiClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
});
|
|
}
|
|
case 'openrouter':
|
|
{
|
|
const keys = resolveApiKeyPool(cfg, 'OPENROUTER_API_KEY');
|
|
if (keys.length === 0) {
|
|
throw new Error(
|
|
'API key required for openrouter. Set OPENROUTER_API_KEY or provide api_key/api_keys in config.',
|
|
);
|
|
}
|
|
return createApiKeyClient(keys, (apiKey) => new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
baseURL: cfg.endpoint ?? 'https://openrouter.ai/api/v1',
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
case 'vercel':
|
|
return new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey: requireApiKey(cfg, 'AI_GATEWAY_API_KEY'),
|
|
baseURL: cfg.endpoint ?? 'https://ai-gateway.vercel.sh/v1',
|
|
});
|
|
case 'zhipuai':
|
|
return new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey: resolveZaiCredential(cfg),
|
|
baseURL: cfg.endpoint ?? 'https://api.z.ai/api/paas/v4',
|
|
});
|
|
case 'xai':
|
|
{
|
|
const keys = resolveApiKeyPool(cfg, 'XAI_API_KEY');
|
|
if (keys.length === 0) {
|
|
throw new Error(
|
|
'API key required for xai. Set XAI_API_KEY or provide api_key/api_keys in config.',
|
|
);
|
|
}
|
|
return createApiKeyClient(keys, (apiKey) => new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
baseURL: cfg.endpoint ?? 'https://api.x.ai/v1',
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
case 'minimax':
|
|
{
|
|
const keys = resolveApiKeyPool(cfg, 'MINIMAX_API_KEY');
|
|
if (keys.length === 0) {
|
|
throw new Error(
|
|
'API key required for minimax. Set MINIMAX_API_KEY or provide api_key/api_keys in config.',
|
|
);
|
|
}
|
|
return createApiKeyClient(keys, (apiKey) => new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
baseURL: cfg.endpoint ?? 'https://api.minimax.io/v1',
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
case 'moonshot':
|
|
{
|
|
const keys = resolveApiKeyPool(cfg, 'MOONSHOT_API_KEY');
|
|
if (keys.length === 0) {
|
|
throw new Error(
|
|
'API key required for moonshot. Set MOONSHOT_API_KEY or provide api_key/api_keys in config.',
|
|
);
|
|
}
|
|
return createApiKeyClient(keys, (apiKey) => new OpenAIClient({
|
|
model: cfg.model,
|
|
apiKey,
|
|
baseURL: cfg.endpoint ?? 'https://api.moonshot.cn/v1',
|
|
}), { cooldownMs: cfg.auth_profile_cooldown_ms });
|
|
}
|
|
case 'bedrock':
|
|
return new BedrockClient({
|
|
model: cfg.model,
|
|
region: cfg.endpoint,
|
|
accessKeyId: cfg.api_key,
|
|
secretAccessKey: cfg.auth_token,
|
|
});
|
|
case 'github':
|
|
return new GitHubModelsClient({
|
|
model: anthropicToGitHubModel(cfg.model) ?? cfg.model,
|
|
apiKey: cfg.api_key,
|
|
endpoint: cfg.endpoint,
|
|
onLoginRequired: async () => {
|
|
const { loginGitHub } = await import('../auth/index.js');
|
|
return loginGitHub((userCode, verificationUri) => {
|
|
console.log(`GitHub login required. Visit: ${verificationUri}`);
|
|
console.log(`Enter code: ${userCode}`);
|
|
});
|
|
},
|
|
});
|
|
case 'synthetic':
|
|
return new SyntheticClient({ model: cfg.model });
|
|
default:
|
|
throw new Error(`Unknown model provider: ${(cfg as Record<string, unknown>).provider}`);
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Map an Anthropic model identifier to its GitHub Models equivalent.
|
|
* Returns undefined if no mapping is known.
|
|
*
|
|
* Anthropic uses hyphens and date suffixes: claude-sonnet-4-5-20250929
|
|
* GitHub Copilot uses dots, no dates: claude-sonnet-4.5
|
|
*/
|
|
export function anthropicToGitHubModel(anthropicModel: string): string | undefined {
|
|
// Explicit mappings for known models
|
|
const MAPPINGS: Record<string, string> = {
|
|
// Sonnet family
|
|
'claude-sonnet-4-20250514': 'claude-sonnet-4',
|
|
'claude-sonnet-4-5-20250929': 'claude-sonnet-4.5',
|
|
'claude-sonnet-4-6-20260217': 'claude-sonnet-4.6',
|
|
// Opus family
|
|
'claude-opus-4-20250514': 'claude-opus-4',
|
|
'claude-opus-4-5-20250918': 'claude-opus-4.5',
|
|
'claude-opus-4-6-20250715': 'claude-opus-4.6',
|
|
// Haiku family
|
|
'claude-3-5-haiku-20241022': 'claude-haiku-4.5',
|
|
'claude-haiku-4-5-20251001': 'claude-haiku-4.5',
|
|
};
|
|
|
|
if (MAPPINGS[anthropicModel]) {return MAPPINGS[anthropicModel];}
|
|
|
|
// Generic fallback: strip date suffix, then convert trailing -N to .N
|
|
// only when preceded by another digit (i.e. "4-5" → "4.5", not "sonnet-5" → "sonnet.5")
|
|
// e.g. "claude-sonnet-4-7-20260301" → "claude-sonnet-4-7" → "claude-sonnet-4.7"
|
|
const dateMatch = anthropicModel.match(/^(.+)-\d{8}$/);
|
|
if (dateMatch) {
|
|
const base = dateMatch[1];
|
|
// Convert "claude-sonnet-4-5" → "claude-sonnet-4.5" (digit-hyphen-digit at end)
|
|
const dotted = base.replace(/(\d)-(\d+)$/, '$1.$2');
|
|
return dotted;
|
|
}
|
|
|
|
return undefined;
|
|
}
|
|
|
|
/**
|
|
* For a given tier config using the Anthropic provider, create a GitHub Models
|
|
* client with the equivalent model as an automatic same-model fallback.
|
|
* Returns undefined if no mapping exists or the tier isn't Anthropic.
|
|
*/
|
|
export function createAutoFallbackClient(tierConfig: { provider: string; model: string }): ModelClient | undefined {
|
|
if (tierConfig.provider !== 'anthropic') {return undefined;}
|
|
|
|
const githubModel = anthropicToGitHubModel(tierConfig.model);
|
|
if (!githubModel) {return undefined;}
|
|
|
|
return new GitHubModelsClient({
|
|
model: githubModel,
|
|
onLoginRequired: async () => {
|
|
const { loginGitHub } = await import('../auth/index.js');
|
|
return loginGitHub((userCode, verificationUri) => {
|
|
console.log(`GitHub login required. Visit: ${verificationUri}`);
|
|
console.log(`Enter code: ${userCode}`);
|
|
});
|
|
},
|
|
});
|
|
}
|
|
|
|
export function createModelRouter(config: Config): ModelRouter {
|
|
const models = config.models;
|
|
|
|
const defaultClient = createClientFromConfig(models.default);
|
|
|
|
const fastClient = models.fast ? createClientFromConfig(models.fast) : undefined;
|
|
const complexClient = models.complex ? createClientFromConfig(models.complex) : undefined;
|
|
const localClient = models.local ? createClientFromConfig(models.local) : undefined;
|
|
|
|
// Build fallback chain — each entry references a tier name or 'local'
|
|
const fallbackChain: ModelClient[] = [];
|
|
for (const providerName of models.fallback_chain) {
|
|
if (providerName === 'local' && localClient) {
|
|
fallbackChain.push(localClient);
|
|
} else if (providerName === 'default') {
|
|
// Allows re-trying the default provider in the chain
|
|
fallbackChain.push(defaultClient);
|
|
} else if (providerName === 'fast' && fastClient) {
|
|
fallbackChain.push(fastClient);
|
|
} else if (providerName === 'complex' && complexClient) {
|
|
fallbackChain.push(complexClient);
|
|
} else if (models.local_providers?.[providerName]) {
|
|
// Named provider from local_providers map
|
|
fallbackChain.push(createClientFromConfig(models.local_providers[providerName]));
|
|
} else {
|
|
logger.warn(`Fallback chain entry "${providerName}" not found — skipping`);
|
|
}
|
|
}
|
|
|
|
// Build per-tier fallbacks from inline fallback config + auto same-model fallbacks.
|
|
// Auto-fallback: when a tier uses Anthropic, automatically insert a GitHub Models
|
|
// client for the same model *before* any user-configured inline fallbacks.
|
|
// This ensures the same model is tried via an alternative provider before
|
|
// degrading to the global fallback chain (which may be a much weaker local model).
|
|
const tierFallbacks = new Map<ModelTier, ModelClient[]>();
|
|
|
|
const tierConfigs: { tier: ModelTier; cfg: typeof models.default | undefined }[] = [
|
|
{ tier: 'default', cfg: models.default },
|
|
{ tier: 'fast', cfg: models.fast },
|
|
{ tier: 'complex', cfg: models.complex },
|
|
{ tier: 'local', cfg: models.local },
|
|
];
|
|
|
|
const autoFallbackTiers: string[] = [];
|
|
for (const { tier, cfg } of tierConfigs) {
|
|
if (!cfg) {continue;}
|
|
|
|
const fallbackList: ModelClient[] = [];
|
|
|
|
// Auto same-model fallback (only when user hasn't configured an inline fallback)
|
|
if (!cfg.fallback) {
|
|
const autoClient = createAutoFallbackClient(cfg);
|
|
if (autoClient) {
|
|
fallbackList.push(autoClient);
|
|
autoFallbackTiers.push(tier);
|
|
}
|
|
}
|
|
|
|
// User-configured inline fallback
|
|
if (cfg.fallback) {
|
|
fallbackList.push(createClientFromConfig(cfg.fallback));
|
|
}
|
|
|
|
if (fallbackList.length > 0) {
|
|
tierFallbacks.set(tier, fallbackList);
|
|
}
|
|
}
|
|
|
|
if (tierFallbacks.size > 0) {
|
|
const tierNames = Array.from(tierFallbacks.keys()).join(', ');
|
|
logger.info(`Per-tier fallbacks configured for: ${tierNames}`);
|
|
}
|
|
if (autoFallbackTiers.length > 0) {
|
|
logger.info(`Auto same-model fallback (via GitHub Models) for: ${autoFallbackTiers.join(', ')}`);
|
|
}
|
|
|
|
logger.info(`Model router: default=${models.default.provider}/${models.default.model}, ` +
|
|
`fallback=[${models.fallback_chain.join(', ')}]`);
|
|
|
|
// Build retry config if enabled
|
|
const retryConfig: RetryConfig | undefined = config.retry.enabled ? {
|
|
maxRetries: config.retry.max_retries,
|
|
initialDelayMs: config.retry.initial_delay_ms,
|
|
backoffMultiplier: config.retry.backoff_multiplier,
|
|
maxDelayMs: config.retry.max_delay_ms,
|
|
nonRetryablePatterns: DEFAULT_RETRY_CONFIG.nonRetryablePatterns,
|
|
} : undefined;
|
|
|
|
if (retryConfig) {
|
|
logger.info(`Retry policy: max_retries=${retryConfig.maxRetries}, initial_delay=${retryConfig.initialDelayMs}ms`);
|
|
}
|
|
|
|
return new ModelRouter({
|
|
default: defaultClient,
|
|
fast: fastClient,
|
|
complex: complexClient,
|
|
local: localClient,
|
|
fallbackChain,
|
|
tierFallbacks,
|
|
retryConfig,
|
|
labels: {
|
|
default: `${models.default.provider}/${models.default.model}`,
|
|
...(models.fast ? { fast: `${models.fast.provider}/${models.fast.model}` } : {}),
|
|
...(models.complex ? { complex: `${models.complex.provider}/${models.complex.model}` } : {}),
|
|
...(models.local ? { local: `${models.local.provider}/${models.local.model}` } : {}),
|
|
},
|
|
});
|
|
}
|