savant-code 0.0.19 → 0.0.21

package/common/src/constants/free-agents.ts

@@ -1,347 +0,0 @@
-import {
-  SAVANT_FREE_GEMINI_PRO_AGENT_IDS,
-  SAVANT_FREE_GEMINI_THINKER_AGENT_ID,
-} from './savant-free-gemini-thinker'
-import {
-  SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID,
-  SAVANT_FREE_DEEPSEEK_V4_PRO_MODEL_ID,
-  SAVANT_FREE_GEMINI_PRO_MODEL_ID,
-  SAVANT_FREE_GLM_V52_MODEL_ID,
-  SAVANT_FREE_HY3_ATLAS_MODEL_ID,
-  SAVANT_FREE_HY3_MODEL_ID,
-  SAVANT_FREE_KIMI_MODEL_ID,
-  SAVANT_FREE_MINIMAX_M3_MODEL_ID,
-  SAVANT_FREE_MIMO_V25_MODEL_ID,
-  SAVANT_FREE_MIMO_V25_PRO_MODEL_ID,
-} from './savant-free-models'
-import { parseAgentId } from '../util/agent-id-parsing'
-
-/**
- * The cost mode that indicates FREE mode.
- * Only allowlisted agent+model combinations cost 0 credits in this mode.
- */
-export const FREE_COST_MODE = 'free' as const
-
-/**
- * The single root agent SavantFree Desktop's hosted (savant-code) harness runs every
- * thread turn under (see savant-free-desktop thread-agent.ts). Unlike the CLI — which
- * has one root id per model (`savant-free-<model>`) — the desktop uses ONE root id
- * for ALL its models, picking the model per tab. It's a first-party free-mode root
- * just like `savant-free*`, so it's listed in SAVANT_FREE_ROOT_AGENT_IDS below; its
- * allowed models are the full desktop picker set (see FREE_MODE_AGENT_MODELS). It
- * carries the "You are Savant" CLI marker in its system prompt so it passes
- * requestHasSavantFreeSystemMarker.
- */
-export const SAVANT_FREE_DESKTOP_THREAD_AGENT_ID = 'savant-free-desktop-thread'
-
-/**
- * Root-orchestrator agent IDs counted as "a savant-free session" for abuse
- * detection and usage auditing. Subagents (scout, basher, etc.) are
- * excluded — they're spawned by the root, so counting them would inflate
- * every user's apparent activity.
- */
-export const SAVANT_FREE_ROOT_AGENT_IDS = [
-  'savant-free',
-  'savant-free-kimi',
-  'savant-free-deepseek',
-  'savant-free-deepseek-flash',
-  'savant-free-mimo-pro',
-  'savant-free-mimo',
-  'savant-free-minimax-m3',
-  'savant-free-glm',
-  // SavantFree Web trial orchestrators (savant_free_bundled_agents.ts). Every root
-  // id in FREE_MODE_AGENT_MODELS that can spawn subagents MUST also be listed
-  // here, or the chat-completions hierarchy gate 403s the subagents with
-  // "Free mode subagents must run under an active savant-free session root"
-  // (2026-07-09 incident: trial runs failed at spawn_agent_inline).
-  'savant-free-hy3',
-  'savant-free-hy3-atlas',
-  SAVANT_FREE_DESKTOP_THREAD_AGENT_ID,
-] as const
-const SAVANT_FREE_ROOT_AGENT_ID_SET: ReadonlySet<string> = new Set(
-  SAVANT_FREE_ROOT_AGENT_IDS,
-)
-
-export const SAVANT_FREE_ROOT_AGENT_ID_BY_MODEL: Record<string, string> = {
-  [SAVANT_FREE_MIMO_V25_PRO_MODEL_ID]: 'savant-free-mimo-pro',
-  [SAVANT_FREE_MIMO_V25_MODEL_ID]: 'savant-free-mimo',
-  [SAVANT_FREE_MINIMAX_M3_MODEL_ID]: 'savant-free-minimax-m3',
-  [SAVANT_FREE_KIMI_MODEL_ID]: 'savant-free-kimi',
-  [SAVANT_FREE_DEEPSEEK_V4_PRO_MODEL_ID]: 'savant-free-deepseek',
-  [SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID]: 'savant-free-deepseek-flash',
-  [SAVANT_FREE_GLM_V52_MODEL_ID]: 'savant-free-glm',
-}
-
-/**
- * After agent roster consolidation (FID-2026-0718-006), all reviewer variants
- * have been merged into the single canonical Verifier agent. The Verifier
- * inherits the parent model via withParentModel(), so no model-specific
- * mapping is needed.
- */
-export const SAVANT_FREE_REVIEWER_AGENT_ID_BY_MODEL: Record<string, string> = {}
-
-export function getSavantFreeRootAgentIdForModel(model: string): string {
-  return SAVANT_FREE_ROOT_AGENT_ID_BY_MODEL[model] ?? 'savant-free'
-}
-
-/**
- * Models that generic subagents may run on when inheriting the parent agent's
- * model. These agents are spawned by free-mode roots, so they must accept any
- * free model the user selected.
- */
-const SAVANT_FREE_SUBAGENT_MODELS = new Set([
-  SAVANT_FREE_MINIMAX_M3_MODEL_ID,
-  SAVANT_FREE_DEEPSEEK_V4_PRO_MODEL_ID,
-  SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID,
-  SAVANT_FREE_KIMI_MODEL_ID,
-  SAVANT_FREE_MIMO_V25_PRO_MODEL_ID,
-  SAVANT_FREE_MIMO_V25_MODEL_ID,
-  SAVANT_FREE_GLM_V52_MODEL_ID,
-  SAVANT_FREE_HY3_MODEL_ID,
-  SAVANT_FREE_HY3_ATLAS_MODEL_ID,
-])
-
-/**
- * Agents that are allowed to run in FREE mode.
- * Only these specific agents (and their expected models) get 0 credits in FREE mode.
- * This prevents abuse by users trying to use arbitrary agents for free.
- *
- * The mapping also specifies which models each agent is allowed to use in free mode.
- * If an agent uses a different model, it will be charged full credits.
- */
-export const FREE_MODE_AGENT_MODELS: Record<string, Set<string>> = {
-  // Root orchestrator
-  'savant-free': new Set([
-    SAVANT_FREE_MINIMAX_M3_MODEL_ID,
-    SAVANT_FREE_DEEPSEEK_V4_PRO_MODEL_ID,
-    SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID,
-    SAVANT_FREE_KIMI_MODEL_ID,
-    SAVANT_FREE_MIMO_V25_PRO_MODEL_ID,
-    SAVANT_FREE_MIMO_V25_MODEL_ID,
-  ]),
-  'savant-free-kimi': new Set([SAVANT_FREE_KIMI_MODEL_ID]),
-  'savant-free-deepseek': new Set([SAVANT_FREE_DEEPSEEK_V4_PRO_MODEL_ID]),
-  'savant-free-deepseek-flash': new Set([
-    SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID,
-  ]),
-  'savant-free-mimo-pro': new Set([SAVANT_FREE_MIMO_V25_PRO_MODEL_ID]),
-  'savant-free-mimo': new Set([SAVANT_FREE_MIMO_V25_MODEL_ID]),
-  'savant-free-minimax-m3': new Set([SAVANT_FREE_MINIMAX_M3_MODEL_ID]),
-  'savant-free-glm': new Set([SAVANT_FREE_GLM_V52_MODEL_ID]),
-  'savant-free-hy3': new Set([SAVANT_FREE_HY3_MODEL_ID]),
-  'savant-free-hy3-atlas': new Set([SAVANT_FREE_HY3_ATLAS_MODEL_ID]),
-
-  // SavantFree Desktop's single hosted root agent — one root id across all its
-  // models (the user picks the model per tab), so it allows the full desktop
-  // picker set. Concurrency is still bounded elsewhere: the free-session
-  // admission gate caps premium-bucket models (incl. MiniMax M3) to one active
-  // session per user (premium_slot_taken), so "one premium model at a time" in
-  // full access holds regardless of this allowlist.
-  [SAVANT_FREE_DESKTOP_THREAD_AGENT_ID]: new Set([
-    SAVANT_FREE_MINIMAX_M3_MODEL_ID,
-    SAVANT_FREE_DEEPSEEK_V4_PRO_MODEL_ID,
-    SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID,
-    SAVANT_FREE_KIMI_MODEL_ID,
-    SAVANT_FREE_MIMO_V25_PRO_MODEL_ID,
-    SAVANT_FREE_MIMO_V25_MODEL_ID,
-    SAVANT_FREE_GLM_V52_MODEL_ID,
-  ]),
-
-  // Generic subagents inherit the parent agent's model, so they must accept
-  // every free model a user might select.
-  scout: SAVANT_FREE_SUBAGENT_MODELS,
-  'file-picker-max': SAVANT_FREE_SUBAGENT_MODELS,
-  'file-lister': SAVANT_FREE_SUBAGENT_MODELS,
-
-  // Research agents
-  'researcher-web': SAVANT_FREE_SUBAGENT_MODELS,
-  'researcher-docs': SAVANT_FREE_SUBAGENT_MODELS,
-
-  // Browser automation
-  'browser-use': SAVANT_FREE_SUBAGENT_MODELS,
-
-  // Command execution
-  basher: SAVANT_FREE_SUBAGENT_MODELS,
-  'tmux-cli': SAVANT_FREE_SUBAGENT_MODELS,
-
-  // Context pruning (spawned inline by the orchestrator)
-  'context-pruner': SAVANT_FREE_SUBAGENT_MODELS,
-
-  // Consolidated Verifier (replaces all code-reviewer-* variants).
-  // The Verifier inherits the parent model via withParentModel().
-  verifier: SAVANT_FREE_SUBAGENT_MODELS,
-
-  // Legacy: kept for the standalone gemini thinker agent if invoked directly.
-  [SAVANT_FREE_GEMINI_THINKER_AGENT_ID]: new Set([
-    SAVANT_FREE_GEMINI_PRO_MODEL_ID,
-  ]),
-}
-
-/**
- * Agents that don't charge credits when credits would be very small (<5).
- *
- * These are typically lightweight utility agents that:
- * - Use cheap models (e.g., Gemini Flash)
- * - Have limited, programmatic capabilities
- * - Are frequently spawned as subagents
- *
- * Making them free avoids user confusion when they connect their own
- * Claude subscription (BYOK) but still see credit charges for non-Claude models.
- *
- * NOTE: This is separate from FREE_MODE_ALLOWED_AGENTS which is for the
- * explicit "free" cost mode. These agents get free credits only when
- * the cost would be trivial (<5 credits).
- */
-export const FREE_TIER_AGENTS = new Set([
-  'scout',
-  'file-picker-max',
-  'file-lister',
-  'researcher-web',
-  'researcher-docs',
-])
-
-export function isSavantFreeRootAgent(fullAgentId: string): boolean {
-  const { publisherId, agentId } = parseAgentId(fullAgentId)
-  if (!agentId) return false
-  if (publisherId && publisherId !== 'savant-code') return false
-  return SAVANT_FREE_ROOT_AGENT_ID_SET.has(agentId)
-}
-
-export function isSavantFreeGeminiThinkerAgent(fullAgentId: string): boolean {
-  const { publisherId, agentId } = parseAgentId(fullAgentId)
-  if (!agentId) return false
-  if (publisherId && publisherId !== 'savant-code') return false
-  return agentId === SAVANT_FREE_GEMINI_THINKER_AGENT_ID
-}
-
-/**
- * True if this agent is permitted to call the premium Gemini Pro model — i.e.
- * one of the two gemini-thinker subagents (CLI `thinker-with-files-gemini` or
- * chat `thinker-gemini`). Publisher-spoof-safe like the other gates: a
- * non-savant-code publisher never matches.
- */
-export function isSavantFreeGeminiProAgent(fullAgentId: string): boolean {
-  const { publisherId, agentId } = parseAgentId(fullAgentId)
-  if (!agentId) return false
-  if (publisherId && publisherId !== 'savant-code') return false
-  return SAVANT_FREE_GEMINI_PRO_AGENT_IDS.has(agentId)
-}
-
-export function shouldUseLocalTokenCountForSavantFreeDeepseekFlash(params: {
-  agentId: string | undefined
-  model: string | undefined
-}): boolean {
-  const { agentId: fullAgentId, model } = params
-  if (!fullAgentId || model !== SAVANT_FREE_DEEPSEEK_V4_FLASH_MODEL_ID) {
-    return false
-  }
-
-  const { publisherId, agentId } = parseAgentId(fullAgentId)
-  if (publisherId && publisherId !== 'savant-code') return false
-  return agentId === 'savant-free-deepseek-flash'
-}
-
-/**
- * Determines whether token counting should use the fast local estimation
- * path instead of the external SavantCode web API. The external API ships
- * the full message history + tools via HTTP on every step, adding serial
- * network overhead (30s timeout × 3 retries) that is unnecessary when:
- * - No SavantCode backend is configured (external/BYOK/OpenCode Go runs)
- * - The model is savant-free-deepseek-flash (existing exception)
- *
- * Local estimation uses gpt-tokenizer with a 1.35× fudge factor — fast
- * and accurate enough for context management. The external API is only
- * needed for SavantCode-hosted paid runs where credit billing requires
- * Anthropic-exact counts.
- */
-export function shouldUseLocalTokenCount(params: {
-  agentId: string | undefined
-  model: string | undefined
-  hasSavantCodeBackend?: boolean
-}): boolean {
-  // Existing narrow exception: savant-free-deepseek-flash always uses local
-  if (shouldUseLocalTokenCountForSavantFreeDeepseekFlash(params)) {
-    return true
-  }
-
-  // If no SavantCode backend is configured, skip the external API entirely
-  // (it would fail with 'Missing SavantCode base URL or API key' anyway)
-  if (params.hasSavantCodeBackend === false) {
-    return true
-  }
-
-  return false
-}
-
-/**
- * Check if a specific agent is allowed to use a specific model in FREE mode.
- * This is the strictest check - validates both the agent AND model combination.
- *
- * Returns true only if:
- * 1. The agent has a valid agent ID
- * 2. The agent is in the allowed free-mode agents list
- * 3. The agent is either internal or published by 'savant-code' (prevents spoofing)
- * 4. The model is in that agent's allowed model set
- */
-export function isFreeModeAllowedAgentModel(
-  fullAgentId: string,
-  model: string,
-): boolean {
-  const { publisherId, agentId } = parseAgentId(fullAgentId)
-
-  // Must have a valid agent ID
-  if (!agentId) return false
-
-  // Must be either internal (no publisher) or from savant-code
-  if (publisherId && publisherId !== 'savant-code') return false
-
-  // Get the allowed models for this agent
-  const allowedModels = FREE_MODE_AGENT_MODELS[agentId]
-  if (!allowedModels) return false
-
-  // Empty set means programmatic agent (no LLM calls expected)
-  // For these, any model check should fail (they shouldn't be making LLM calls)
-  if (allowedModels.size === 0) return false
-
-  // Exact match first
-  if (allowedModels.has(model)) return true
-
-  // OpenRouter may return dated variants (e.g. "minimax/minimax-m3-20260211")
-  // so also check date-like suffixes. Do not accept arbitrary suffixes:
-  // "mimo-v2.5-pro" must not match the non-pro "mimo-v2.5" allowlist entry.
-  for (const allowed of allowedModels) {
-    const prefix = allowed + '-'
-    if (model.startsWith(prefix)) {
-      const suffix = model.slice(prefix.length)
-      if (/^\d{6,8}(?:$|[-:])/.test(suffix)) return true
-    }
-  }
-
-  return false
-}
-
-/**
- * Check if an agent should be free (no credit charge) for small requests.
- * This is separate from FREE mode - these agents get free credits only
- * when the cost would be trivial (<5 credits).
- *
- * Handles all agent ID formats:
- * - 'scout'
- * - 'scout@1.0.0'
- * - 'savant-code/scout@0.0.2'
- */
-export function isFreeAgent(fullAgentId: string): boolean {
-  const { publisherId, agentId } = parseAgentId(fullAgentId)
-
-  // Must have a valid agent ID
-  if (!agentId) return false
-
-  // Must be in the free tier agents list
-  if (!FREE_TIER_AGENTS.has(agentId)) return false
-
-  // Must be either internal (no publisher) or from savant-code
-  // This prevents publisher spoofing attacks
-  if (publisherId && publisherId !== 'savant-code') return false
-
-  return true
-}

package/common/src/constants/gemini.ts

@@ -1,6 +0,0 @@
-/** Stable OpenRouter model ID for lightweight Gemini agent workloads. */
-export const GEMINI_3_1_FLASH_LITE_MODEL_ID = 'google/gemini-3.1-flash-lite'
-
-/** Retired model ID emitted by SavantFree clients released before July 2026. */
-export const LEGACY_GEMINI_3_1_FLASH_LITE_PREVIEW_MODEL_ID =
-  'google/gemini-3.1-flash-lite-preview'

package/common/src/constants/grant-priorities.ts

@@ -1,13 +0,0 @@
-import type { GrantType } from '@savant-code/common/types/grant'
-
-// Lower = consumed first
-export const GRANT_PRIORITIES: Record<GrantType, number> = {
-  subscription: 10,
-  free: 20,
-  referral_legacy: 30, // Legacy recurring referrals (renews monthly, consumed first)
-  ad: 40,
-  referral: 50, // One-time referrals (never expires, preserved longer)
-  admin: 60,
-  organization: 70,
-  purchase: 80,
-} as const

package/common/src/constants/hosts.ts

@@ -1,6 +0,0 @@
-/**
- * Prod origin of the savant-free.com web app (marketing site + browser sign-in).
- * The CLI and SavantFree Desktop both send users here for the device-code login
- * flow; sharing the literal keeps their prod defaults from drifting.
- */
-export const SAVANT_FREE_WEB_URL_PROD = 'https://savant-free.com'

package/common/src/constants/images.ts

@@ -1,53 +0,0 @@
-/**
- * Image-related constants shared across the codebase
- */
-
-/**
- * Extension to MIME type mapping for supported image formats
- */
-export const IMAGE_EXTENSION_TO_MIME: Record<string, string> = {
-  '.jpg': 'image/jpeg',
-  '.jpeg': 'image/jpeg',
-  '.png': 'image/png',
-  '.webp': 'image/webp',
-  '.gif': 'image/gif',
-  '.bmp': 'image/bmp',
-  '.tiff': 'image/tiff',
-  '.tif': 'image/tiff',
-}
-
-/**
- * Supported image extensions (derived from IMAGE_EXTENSION_TO_MIME)
- */
-export const SUPPORTED_IMAGE_EXTENSIONS = new Set(
-  Object.keys(IMAGE_EXTENSION_TO_MIME),
-)
-
-/**
- * Check if a file extension is a supported image format
- */
-export function isSupportedImageExtension(ext: string): boolean {
-  return SUPPORTED_IMAGE_EXTENSIONS.has(ext.toLowerCase())
-}
-
-/**
- * Get MIME type for an image extension
- */
-export function getImageMimeType(ext: string): string | null {
-  return IMAGE_EXTENSION_TO_MIME[ext.toLowerCase()] ?? null
-}
-
-/**
- * Image extensions as a regex alternation pattern (without dots)
- * e.g., "jpg|jpeg|png|webp|gif|bmp|tiff|tif"
- */
-export const IMAGE_EXTENSIONS_PATTERN = Object.keys(IMAGE_EXTENSION_TO_MIME)
-  .map((ext) => ext.slice(1)) // Remove leading dot
-  .join('|')
-
-// Size limits for image uploads
-// Research shows Claude/GPT-4V support up to 20MB, but we use practical limits
-// for good performance and token efficiency
-export const MAX_IMAGE_FILE_SIZE = 10 * 1024 * 1024 // 10MB - allow larger files since we can compress
-export const MAX_IMAGE_BASE64_SIZE = 1 * 1024 * 1024 // 1MB max for base64 after compression
-export const MAX_TOTAL_IMAGE_SIZE = 5 * 1024 * 1024 // 5MB total for multiple images

package/common/src/constants/index.ts

@@ -1,8 +0,0 @@
-// Re-export all constants from domain-specific files for backwards compatibility
-// This allows existing imports from '@savant-code/common/old-constants' to continue working
-
-export * from './model-config'
-export * from './limits'
-export * from './ui'
-export * from './paths'
-export * from './chatgpt-oauth'

package/common/src/constants/knowledge.ts

@@ -1,46 +0,0 @@
-import path from 'path'
-
-/**
- * The primary/default knowledge file name.
- * Used when creating new knowledge files.
- */
-export const PRIMARY_KNOWLEDGE_FILE_NAME = 'knowledge.md'
-
-/**
- * Knowledge file names in priority order (highest priority first).
- * Used for both project knowledge files and home directory user knowledge files.
- */
-export const KNOWLEDGE_FILE_NAMES = [
-  PRIMARY_KNOWLEDGE_FILE_NAME,
-  'AGENTS.md',
-  'CLAUDE.md',
-] as const
-
-/**
- * Pre-computed lowercase knowledge file names for efficient matching.
- */
-export const KNOWLEDGE_FILE_NAMES_LOWERCASE = KNOWLEDGE_FILE_NAMES.map((name) =>
-  name.toLowerCase(),
-)
-
-/**
- * Checks if a file path is a knowledge file.
- * Matches:
- * - Exact file names: knowledge.md, AGENTS.md, CLAUDE.md (case-insensitive)
- * - Pattern: *.knowledge.md (e.g., authentication.knowledge.md)
- */
-export function isKnowledgeFile(filePath: string): boolean {
-  const fileName = path.basename(filePath).toLowerCase()
-
-  // Check for exact matches with standard knowledge file names
-  if (KNOWLEDGE_FILE_NAMES_LOWERCASE.includes(fileName)) {
-    return true
-  }
-
-  // Check for *.knowledge.md pattern (e.g., authentication.knowledge.md)
-  if (fileName.endsWith('.knowledge.md')) {
-    return true
-  }
-
-  return false
-}

package/common/src/constants/limits.ts

@@ -1,23 +0,0 @@
-export const PROFIT_MARGIN = 0.055
-
-export const REQUEST_CREDIT_SHOW_THRESHOLD = 1
-export const MAX_DATE = new Date(86399999999999)
-export const BILLING_PERIOD_DAYS = 30
-export const SESSION_MAX_AGE_SECONDS = 30 * 24 * 60 * 60 // 30 days
-export const SESSION_TIME_WINDOW_MS = 30 * 60 * 1000 // 30 minutes - used for matching sessions created around fingerprint creation
-// New SavantCode accounts receive a one-time free credit grant on signup.
-export const SIGNUP_FREE_CREDITS_GRANT = 500
-// Signup credits are limited to GitHub-first users whose account is older than
-// this many calendar months. Google-first signups receive no signup credits.
-export const SIGNUP_CREDIT_MIN_GITHUB_ACCOUNT_AGE_MONTHS = 4
-
-// New accounts do not receive monthly free credits; grandfathered monthly grants
-// are based on previous expiring free grants instead of this default.
-export const DEFAULT_FREE_CREDITS_GRANT = 0
-
-// Credit pricing configuration
-export const CREDIT_PRICING = {
-  CENTS_PER_CREDIT: 1, // 1 credit = 1 cent = $0.01
-  MIN_PURCHASE_CREDITS: 100, // $1.00 minimum
-  DISPLAY_RATE: '$0.01 per credit',
-} as const