LyDos hangi sorunu çözüyor?

LyDos, kurumsal AI operasyonlarını tek kontrol düzleminden yönetir: 12 LLM sağlayıcı arasında akıllı yönlendirme, 24/7 otonom kod mühendisliği, KSL cihaz imzası ve ASR hash-zincirli denetim. Sağlayıcı kilitlemesini ortadan kaldırır, AI gözetim/uyum gereksinimlerini karşılar.

Hangi LLM sağlayıcıları destekleniyor?

Groq, Anthropic, OpenAI, Mistral, Gemini, DeepSeek, Z.AI, Qwen, Cohere, NVIDIA NIM, Kimi ve Ollama dahil 12 sağlayıcı. Tek API üzerinden otomatik fallback, maliyet/gecikme tabanlı routing ve sağlayıcı-özel özellik bayrakları.

MCP (Model Context Protocol) entegrasyonu nasıl?

162 üretim MCP aracı entegre. Claude Code, Cursor, Cline, OpenAI Codex CLI, Aider ve VS Code Continue dahil 6 AI CLI ile uyumlu. Tüm tool çağrıları KSL imzalı ve ASR'ye zincirli loglanır.

Sovereign güvenlik yığını ne içeriyor?

13 üretim sovereign motoru: DPI (çoklu-düğüm karar), LSSA (10 katman güvenlik), KSL (cihaz imzası, özel anahtar sunucuda tutulmaz), TUP (güvenilir güncelleme), ASR (hash-zincirli denetim), LSIA (bağışıklık + veto), GCI (gizlilik koruyucu örüntü konsensüsü) ve diğerleri.

Veri egemenliği ve uyum?

EU AI Act hazır. HIPAA BAA-uygun cluster canlı (SaMD scope hariç). GDPR/KVKK destekli. Tüm kullanıcı verisi per-user encrypted vault'ta. PG :5434 federation tek doğruluk kaynağı (sessiz SQLite fallback yok).

Kurulum ne kadar sürer?

Tek satır: curl -fsSL https://lydos.ailydian.com/install | sh — 15 saniyede çalışır kurulum. CLI, MCP server, varsayılan sağlayıcı yapılandırması ve yerel ajan rejistrisi otomatik gelir.

LyDos

Slack

Trust Center · Şeffaflık

Bugün gerçek · Yol haritasında

Name: LyDos Orchestration Platform
Author: LYDOS

Sprint 12 denetimi pazarlama iddiası ile üretim gerçekliği arasında bir sapma yakalamıştı (170–230 ajan iddiası, 120 gerçek). Bu bölüm tekrarını engeller: her satır canlı sevkiyat dosyasına, denetim raporuna, ya da manifest/canonical-stats.json anahtarına bağlı.

Transparency

Real today, on the roadmap

Sprint 12 audit caught drift between marketing claims and production reality (170-230 agent claim vs 120 actual). This section prevents that — every line is verifiable against the live manifest or audit report.

Real today (in production)

120 agents (13 Python + 107 YAML) — loaded at startupmanifest/canonical-stats.json:agents_total
13 sovereign engines (DPI/LocalSec/KSL/TUP/GCE/ASR/DSL/LSIA/GCI…) live PG :543413/13 /health endpoints respond
KSL signing on 100% of critical routes (34/34 audit-verified)compliance/audit/sprint-12/AUDIT-REPORT.md
ASR hash-chained audit chain (208+ live records, 5 tables)asr_decisions table on PG :5434
RLS isolation 22/22 tests PASS — multi-tenant data separation proventests/test_rls_isolation.py
12 LLM providers per-user vault (AES-256-GCM) + Cmd+K palettecore/integrations/llm_provider_integration.py
Single VPS production deployment — controlled betadeploy/vps/lydos-*.service

On the roadmap

Multi-region + HPA (autoscaling) — currently single VPS, code not yet shippedcapability_runtime_status.multi_region: code_only
FedRAMP authorisation — Phase 0 prep, NOT authorisedmanifest §audit_history
FDA SaMD scope — EXCLUDED; only BAA-eligible HIPAA cluster is liveaudit-verified scope boundary
7 locales fully localised, 23 locales planned (Q3/Q4 2026)manifest §i18n_locales
Truthful homepage rewrite + /settings live wiring + /integrations enterprise search — in progress (Frontend Overhaul FO-3..9)compliance/audit/frontend-overhaul/FRONTEND_OVERHAUL_RFC.md

How this list stays honest

Manifest is the source. Counts come from manifest/canonical-stats.json through the typed Python loader (core/integrations/canonical_stats.py) and the TypeScript mirror (web/src/lib/canonical-stats.ts). The loader fails to import if the manifest disagrees with the live LLM registry, so a future drift gets caught at backend boot.
Runtime status is explicit. Every capability has a runtime_status field — code_only, deployed_partial, or deployed_full— so “code exists in repo” never gets passed off as “deployed in production.”
CI drift gates. tests/test_sprint37_phase2_canonical_stats_consumer.py scans marketing pages for hard-coded counts that should reference the canonical manifest; the build fails if anyone re-introduces a literal “120 agents” string.

Canonical numbers →Compliance frameworks →Sovereignty architecture →

How this list stays honest

Manifest is the source. Counts come from manifest/canonical-stats.json through the typed Python loader (core/integrations/canonical_stats.py) and the TypeScript mirror (web/src/lib/canonical-stats.ts). The loader fails to import if the manifest disagrees with the live LLM registry, so a future drift gets caught at backend boot.

Runtime status is explicit. Every capability has a runtime_status field — code_only, deployed_partial, or deployed_full— so “code exists in repo” never gets passed off as “deployed in production.”

CI drift gates. tests/test_sprint37_phase2_canonical_stats_consumer.py scans marketing pages for hard-coded counts that should reference the canonical manifest; the build fails if anyone re-introduces a literal “120 agents” string.