SAGA — Sovereign Communication Integrity Layer
SAGA - Sovereign Communication Integrity LayerLocal audit proxy that scores every assistant turn for continuity, register, hedging, and standing-instruction compliance. Composes with FREYA, BRAGI, MIMIR.What it does: SAGA sits between any LLM client and any provider on TCP 7800 (or stacks behind MIMIR on 7799). Captures both sides of the conversation, runs six audit checks, attaches a Continuity Score 0-100 plus structured flags to every response.Six audit checks: state_drift: contradiction across turns - register_mismatch: formal/casual/technical alignment with user - false_certainty: confidence proportional to evidence (reads MIMIR verdict headers when present) - ignored_context: turn ignores enumerated user constraints - standing_instruction_violation: explicit user rule violated - ethics_flag: medical/legal/financial/self-harm awareness Install: Run SAGA-Setup-v0.1.0.exe. First launch surfaces a 3-question calibration wizard. Defaults to 127.0.0.1:7800. No telemetry. No phone-home. Local-only.Tiers: Personal $149/yr - 1 seat, T1+T2 storage tiers - Business $599/yr - 5 seats, T1+T2+T3, audit export - Enterprise $2,499/yr - unlimited seats, custom patterns, compliance reporting, SSO Composition: client -> MIMIR :7799 -> SAGA :7800 -> upstream LLM. MIMIR audits WHAT (truth, knowledge integrity). SAGA audits HOW and WHEN (register, continuity, context-fit, drift). Headers cross between them.Requirements: Windows 10/11. Python embedded runtime bundled in the installer (~7MB exe). Optional: Ollama with qwen2.5:7b for higher-quality state tracking and register classification (lexical fallback works without it).Sovereignty: All data stays on the user machine. T3 cross-session storage is opt-in and DPAPI-encrypted at rest on Windows. Each install starts fresh. Adaptive learning is per-machine and never synced.Snake River AI - snakeriverai.com/saga
Get it → clintwave84.gumroad.com