EvaBot Docs

CHANGELOG

время чтения ~8 мин.

Changelog

All notable changes to EvaBot Online will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[v0.1.0] - 2026-09-07 — Cyber-Terminal TUI, Frontier 3.8 Fleet, Accounting & Agent Factory

🖥️ Cyber-Terminal & TUI Isomorphism

  • Pure CRT Monospace TUI: Eliminated bloated cards and borders; strict 16px monospace layout in dark theme (#0a0a0a).
  • 5-Line System Header: Line 1 (Online + 1s Real-time Ping + Mesh RTT + Pulse Wave), Line 2 (Active Model + Mode + 78 Models Pool), Line 3 (Commands Bar), Line 4 (Databases OK), Line 5 (ASCII Load Bars CPU/RAM + Heartbeat).
  • Parity Across All Interfaces: Browser, Node.js CLI (terminal-chat.ts), text browsers (Lynx/w3m), and curl http://127.0.0.1:3000/.
  • Autoscroll Support: Automatic smooth scrolling to bottom during streaming chunks and message additions.

🤖 Frontier Fleet & Google Ecosystem

  • **Gemini 3.8 Flash as Default 1: Frontier 1M-context model with 100% Free Quota set as primary engine.
  • Search Grounding Guaranteed: Direct live web fact retrieval via Vertex AI Google Search Grounding with zero extra cost.
  • Cascading Fallback Chain: Strict ranked fallback: gemini-3.8-flash ➔ gemini-3.1-pro ➔ gemini-3.1-flash ➔ deepseek-r1:free ➔ qwen-2.5-coder-32b:free.

💰 Accounting & Financial Ledger (/cost)

  • AccountingEngine: Tracks live token consumption, input/output cost, and calculated savings against commercial flagships (0.2250 - 0.3000 saved per task).
  • Hardware OpEx Itemization: Frankfurt Compute Core (178.40), Iowa Edge Ingress (7.14), Disks (12.00), Mesh (5.00), Subscriptions (Google AI Pro 20, Colab Pro 10, OpenRouter 25) = 257.54/mo ($0.3577/h).
  • Zero-Cost Unit Economics: Creation of autonomous agents on Gemini 3.8 Flash / 3.1 Pro is verified at $0.00.

🏢 Agent Constructor (/company [free|paid])

  • AgentBuilder: Generates 10-agent autonomous corporations with defined missions, assigned models, and required MCP/LSP tools.
  • 100% Free Fleet: 10 specialized agents running on 100% free models.
  • Commercial Frontier Fleet: 10 specialized agents running on Claude 3.7 Sonnet, OpenAI o1, GPT-4o, Codestral 2501, etc.
  • Model Technical Passports (/info <model>): Full specifications, token costs, context limits, and composite benchmark ratings.

🔌 MCP (21 Servers) & LSP (4 Servers) Integration

  • Commands /mcp & /lsp: Live status reporting of all 21 MCP servers and 4 Language Servers in PATH (TypeScript, Python, HTML/CSS/JSON, Markdown).

🧪 100% Test Coverage

  • 13 Test Suites: All 13 suites (npm test) pass with 100% success (ModelTests, ChatTests, ServerTests, CoreEngineTests, UniversalClientTests, ConsiliumTests, RolesTests, AnsiStreamEngineTests, PluginManagerTests, EventBusTests, LLMProvidersTests, KnowledgeBaseTests, ConsiliumNewTests, AccountingAndBuilderTests).

[v0.0.2] - 2026-09-07 — Refactored & Hardened

🛡️ Added - Security

  • IP blocking system with 8 pre-blocked malicious IPs (45.148.10.9, 43.157.188.74, etc.)
  • Rate limiting middleware (100 req/min per IP)
  • 17 regex patterns for suspicious path detection (WordPress/.env/admin/etc)
  • Auto-block mechanism: 20 suspicious requests → 24h ban
  • Security endpoints: GET/POST /api/security/{status,report,block,unblock}

📚 Added - Knowledge Base

  • EvaLine Knowledge Base integration: 182 documents loaded from evaline-com-ua
  • 6 languages supported: EN, UK, RU, PL, RO, DE
  • 4 backend types: memory (active), json, sqlite, vector
  • Full-text search with relevance scoring
  • /kb command for terminal-style control
  • KB endpoints: GET/POST /api/kb/{status,search,list,backend,command}

🚨 Added - Alerting System

  • Multi-channel alerts: console, file, webhook, email, syslog, desktop
  • 4 severity levels: low, medium, high, critical
  • Rate limiting (60s cooldown per alert type)
  • Alert endpoints: GET/POST /api/alerts/{stats,send,channel,config}
  • Auto-integration with Security module
  • Environment variable config: ALERT_WEBHOOK_URL, ALERT_EMAIL_TO, SYSLOG_HOST

📊 Added - Comprehensive Logging

  • 12 log categories: SYSTEM, HTTP, USER, LLM, MODEL, KB, STORAGE, AUTH, PROCESS, DIAG, CLI, BROWSER
  • 3 log files: evabot.log, user-actions-{date}.log, errors-{date}.log
  • HTTP request logging with IP, User-Agent, duration
  • In-memory buffer (1000 entries) for fast access
  • Log endpoints: GET /api/logs/{files,read,recent}

🏗️ Changed - Refactoring

  • server.ts refactored from 815 → 211 lines (-74%)
  • Modular router architecture: 7 separate route files
  • Router.ts base class with RouteContext + withErrorHandling wrapper
  • ChatRouter (89 lines) - chat, stream, consilium, roles
  • ModelsRouter (55 lines) - models, top, free, paid
  • KbRouter (70 lines) - KB endpoints
  • LogsRouter (30 lines) - logs endpoints
  • SecurityRouter (39 lines) - security endpoints
  • AlertsRouter (41 lines) - alerts endpoints

🐛 Fixed - Critical Bugs

  • ConsiliumEngine.ts was BROKEN (12 TypeScript errors)
    • Added ModelRegistry.estimateTokens() method
    • Added ModelRegistry.calculateCost() method
    • Added ModelRegistry.getTop10FreeModels() method
    • Added ModelRegistry.getTop10PaidSmartestModels() method
    • Added TokenCostEstimate interface
    • Result: 0 TypeScript errors (was 12)
  • UniversalLlmClient.ts - removed invalid tier === 'OpenRouter Paid' check
  • KB JSON.stringify was crashing on Set serialization
    • Replaced recursive getBackendDescription() with static lookup
    • Fixed Set → Array.from() conversion

🗑️ Removed

  • dist/ from Git (1.3MB, 114 files) - now in .gitignore
  • legacy_archive/ (17MB, 32 files) - old code, now in .gitignore
  • src/plugins/voice/ (5 files) - unused voice plugin code
  • src/web/voice/ (5 files) - unused voice UI code
  • .env.bak - backup file with potential secrets

📁 Added

  • config/Caddyfile - production reverse proxy with WAF, rate limits, security headers
  • config/fail2ban-filter.conf - fail2ban filter for EvaBot logs
  • config/fail2ban-jail.conf - 2 jails (attack + rate)
  • .github/workflows/deploy.yml - CI/CD for 2-server monorepo deploy
  • docs/security/SECURITY_AUDIT.md - full security audit report
  • docs/models/MODELS_CATALOG.md - separated free/paid models
  • docs/deployment/MONOREPO.md - monorepo documentation
  • docs/development/v0.0.1-IMPLEMENTATION.md - v0.0.1 implementation report

🔒 Security

  • Detected and blocked 432 WordPress exploit attempts from IP 45.148.10.9 (Techoff SRV, NL)
  • Blocked CVE-2024-31210 (Batch RCE, CVSS 9.8)
  • Blocked CVE-2024-32336 (Gravity SMTP LFI, CVSS 7.5)
  • 8 unique attacker IPs identified from NL, BR, US, DE, TH, SG
  • 0 successful attacks (WordPress not installed)

[v0.1.0] - 2026-09-07 — Plugin Architecture & Multi-Provider Gateway

🆕 Added - Plugin System

  • Plugin Architecture - автономные плагины с динамической загрузкой
  • PluginManager - singleton с поддержкой enable/disable, dependencies, health checks
  • PluginEventBus - pub/sub для межплагинного взаимодействия
  • Plugin interface - manifest, initialize(), shutdown(), healthCheck()
  • Plugin routes & commands - каждый плагин может регистрировать API routes и CLI commands

🆕 Added - Plugins

  • LLMProvidersPlugin - единый gateway для 5 провайдеров:
    • Google AI (Gemini) - default, free quota
    • OmniRoute (Internal LiteLLM Proxy) - daemon cluster
    • OpenRouter - 78 моделей, free tier
    • OpenCode Go - код-инференс
    • KiloCode - free models aggregator
  • ConsiliumPlugin - Multi-Agent Engine (4 режима: solo, broadcast, dialogue, consilium)
  • KnowledgeBasePlugin - EvaLine KB (178 документов, 6 языков)

🆕 Added - API Endpoints

  • GET /api/plugins - список всех плагинов
  • GET /api/plugins/health - health check всех плагинов
  • GET /api/plugins/:id - информация о плагине
  • POST /api/plugins/:id/enable - включить плагин
  • POST /api/plugins/:id/disable - выключить плагин
  • GET /api/llm/providers - статус всех LLM провайдеров
  • POST /api/llm/chat - чат через любой провайдер
  • POST /api/llm/test - тест провайдера

🆕 Added - Visualizer

  • /visualize.html - интерактивная визуализация всей системы
  • Live метрики, plugins, models, security, logs
  • Архитектурные диаграммы (User → EvaFace → EvaBrain → 4 Providers)
  • Flow diagrams (request, consilium, KB)
  • Real-time updates каждые 10 секунд

🆕 Added - Tests

  • 13 test suites, 100% pass rate
  • tests/plugin-manager.test.ts - 12 tests for PluginManager + EventBus
  • tests/llm-providers.test.ts - 19 tests for LLMProvidersPlugin
  • tests/knowledge-base.test.ts - 20 tests for KnowledgeBasePlugin
  • tests/consilium-new.test.ts - 12 tests for ConsiliumPlugin
  • Total: 60+ test assertions, all passing

🔧 Changed

  • server.ts - интегрирован PluginManager, plugin routes в router
  • TypeScript errors - 0 (clean compile)
  • Endpoints - все /api/kb/* переехали в KnowledgeBasePlugin
  • Endpoints - все /api/consilium/* переехали в ConsiliumPlugin

📦 Plugin Architecture Benefits

  • Modularity - каждый компонент в своем плагине
  • Hot-swap - enable/disable без перезапуска
  • Dependencies - автоматическая проверка зависимостей
  • Health monitoring - каждый плагин имеет healthCheck
  • Easy testing - каждый плагин тестируется отдельно
  • Extensibility - легко добавить новый плагин

[v0.0.1] - 2026-09-03 — MVP Release

Added

  • Initial cyber-terminal interface (TUI + Web)
  • UniversalLlmClient with 78 models
  • Google Gemini, OmniRoute, OpenRouter, OpenCode providers
  • Basic ModelRegistry with 78 models
  • Chat endpoint (POST /api/chat)
  • Streaming chat (POST /api/chat/stream)
  • Consilium engine (4 modes: solo, broadcast, dialogue, consilium)
  • Boot diagnostics
  • Cluster monitor (Frankfurt + Iowa)
  • Health check endpoint
  • Worklog API (TSV/LOG/TXT/MD formats)

[Unreleased] - Planned

v0.1.0

  • Vector embeddings (Gemini embedding-004)
  • ChromaDB integration
  • Real semantic search
  • Frontend mobile-optimized UI
  • localStorage chat history
  • Code highlighting
  • Copy buttons

v0.2.0

  • Consilium v2 with 10+ agents
  • Voting system
  • Consensus arbiter improvement
  • WebSocket real-time chat

v0.3.0

  • Voice input/output
  • Audio streaming
  • Chat export (PDF, Markdown)
  • Multi-language UI improvements

v0.4.0

  • Mobile app (PWA)
  • OAuth2 authentication
  • Multi-user sessions
  • Usage analytics
  • Billing dashboard

Format: Keep a Changelog
Versioning: Semantic Versioning
Status: Active development

[v0.2.0] - 2026-09-08 — Modular Architecture & Security Hardening

🏗️ Added

  • 12 modular repos: eva-brain, eva-voice, eva-face, eva-db, eva-history, eva-memory, evaline-consilium, evabot-server, eva-server, evaline-server, eva-docs, eva-reports
  • eva-docs: unified documentation hub (architecture, ops, agents, domains, models, roadmap, security)
  • eva-reports: chronological session reports with indexing
  • Branch protection on all active repos’ main branches (PR required + 1 approval)
  • GitHub Actions CI/CD: automated build → sync-modules → deploy → smoke test
  • sync-modules.sh: keeps monorepo and module repos in sync
  • Watchdog timer: crash-loop detection with threshold alerting via systemd timer (5min interval)
  • Uptime monitor: cron-based health checks logging to /var/log/evabot-uptime.log

🔒 Changed / Fixed

  • GitHub token hygiene: removed embedded tokens from all git remote URLs (ghp_*); switched to gh auth credential helper
  • GCP key redaction: removed leaked API key from REPORT.md (GitHub Push Protection)
  • google-calendar MCP port conflict: changed PORT 3000 → 3800 in opencode/kilo configs, resolves brain crash-loop
  • Removed release tarballs from git history; added backups/ to .gitignore
  • CompanyKnowledge sources → ../eva-reports/reports/ (single source of truth)
  • docs-site/content → symlink → eva-docs (eliminated content duplication)
  • MANIFESTO.md → symlink (deduplicated from public/)
  • Single-viewport dashboard: evaline.network zero-scroll layout (mobile-first 320px → desktop)

в этой папке 1 элемент