CHANGELOG
What's New
Every release, every fix, every improvement — documented transparently.
v3.2.0
CurrentJune 9, 2026- AddedExogram Knowledge Graph — interactive force-directed visualization of the entire semantic trust network with custom canvas rendering, confidence-based color coding, and click-to-inspect detail panels
- AddedBehavioral Synthesis Engine — AI-powered cross-entry inference layer that extracts behavioral profiles, dominant themes, temporal shifts, contradictions, and predictive insights
- AddedFull-density graph rendering — all ledger entries now appear in the graph (plain-text claims auto-map to User → states → Claim nodes)
- AddedGraph search, zoom controls, and full-screen exploration mode
- SecurityUUID validation hardening across all API endpoints — prevents 500 errors from malformed identifiers (EXOGRAM-3W)
- ImprovedDashboard sidebar version bumped to v3.2
- ImprovedInternal architecture references unified under Exogram branding
- ImprovedREADME updated with Knowledge Graph, Behavioral Synthesis, and UUID hardening documentation
v3.1.0
May 17, 2026- SecurityCryptographic audit chain — every audit entry SHA-256 chained to its predecessor, creating a verifiable append-only log for Enterprise-grade / EU AI Act compliance
- AddedAgent Identity Resolution (Gate 0) — per-agent permit/deny lists with prefix wildcard matching, enforced before policy evaluation
- AddedGraph Context Enrichment — Layer 2 semantic environment (entity anchoring, graph disambiguation, session ledger) now wirable into production /v2/evaluate via enrich_context flag
- AddedAgent identities table with RLS, unique constraints, and auto-updating triggers
- Added12 Architecture Deep-Dive pages — environment-centric thesis, semantic gravity, event-sourced ledger, runtime compiler, and 8 more
- ImprovedAudit chain module consolidated — evaluate.py and commit.py now share a single cryptographic audit writer (audit_chain_v2.py)
- ImprovedSitemap expanded from 319 to 331+ indexed URLs with architecture hub-and-spoke
- Improvedllm.txt, llms.txt, llms-full.txt updated with architecture deep-dive scope and canonical Exogram concepts
v3.0.0
March 26, 2026- AddedExogram Context Construction Engine (Layer 2) — deterministic semantic-to-graph translation
- AddedExecution Idempotency Engine — mathematically seals transactions against double-spend network retries via HTTP 409 Locks
- AddedSemantic Conflict Resolution — evaluates and blocks mutated facts lacking explicit 'supersedes' consensus
- SecurityGhost Vector Elimination — fully syncs Supabase ledger archival with Pinecone vector tombstones to kill semantic drift
- SecurityCircular Graph Prevention — strict Postgres recursive CTE triggers block infinite evaluation topology loops
- ImprovedProving Ground UI updated with 3-tab static boundary demonstrations (Conflict Resolution & Idempotency)
v2.4.0
March 23, 2026- AddedAI Governance Glossary — 27 indexed term pages with DefinedTerm JSON-LD (/glossary)
- AddedLearning Hub — 10 educational articles (What Is AI Governance?, Prompt Injection, Guardrails, etc.) with FAQ schema (/learn)
- AddedProduction Readiness Analyzer — paste your agent code, see your risk exposure (/analyze)
- Added18 individual comparison routes — Exogram vs OpenAI, Anthropic, LangChain, CrewAI, and 14 more (/compare)
- Added5 integration deep-dive pages — MCP, ChatGPT, Chrome Extension, REST API, LangChain (/integrations)
- AddedBlog launch — 'Why We Built Zero Trust for AI Execution' (/blog)
- AddedRSS feed — /feed.xml for blog syndication
- AddedOG images — branded social cards for all major pages
- ImprovedSitemap expanded to 70+ indexed URLs — glossary, learn, integrations, comparisons, blog, feeds
- Improvedrobots.txt updated with RSS + humans.txt discovery
- Improvedllms-full.txt refreshed with Zero Trust positioning and 0.07ms/137 RPS metrics
- ImprovedProtocol and RFC pages updated with production metrics (0.07ms deterministic enforcement, 137 RPS)
- ImprovedPricing layout refreshed with Zero Trust branding and enriched Product JSON-LD
- SecurityInternal linking hardened across all new pages for topical authority
v2.3.0
March 19, 2026- ImprovedUpgraded LLM inference from Gemini 2.5 Flash to Gemini 3.1 Flash Lite — faster responses, lower latency
- ImprovedUpgraded embedding model from gemini-embedding-001 to Gemini Embedding 2 Preview — multimodal (text, image, audio, PDF) in a unified vector space
- AddedPhase 2 telemetry — per-request compute_latency_ms, agent_id, and raw_intent captured across all critical endpoints
- AddedThe Exogram Manifesto — 'The End of Outsourced AI Safety' now live at /manifesto with full Magic UI design and JSON-LD schema
- AddedThe 2030 Vision — 'Autonomous AI Governance' now live at /vision with three-phase trajectory, data monopoly thesis, and structured data
- AddedEAAP RFC-0001 published to GitHub — open specification for community review and contribution
- ImprovedAll 7 backend files updated for model migration — llm_router, llm_router_hardened, llm_service, MCP server, main API
- ImprovedAudit log enrichment — /v2/evaluate, /ledger/propose, and /vault/context now persist latency and intent telemetry
- ImprovedMultimodal embedding support unlocked — embed_multimodal() now uses latest Gemini Embedding 2 model natively
- ImprovedSite navigation updated — Manifesto and Vision added to header nav across all pages
v2.2.0
March 17, 2026- SecurityRemoved authentication fallbacks (X-User-ID header, DEFAULT_USER_ID) — all vault endpoints now require proper auth
- SecurityPurged all hardcoded JWT secrets — JWT_SECRET_KEY is now mandatory, app crashes on missing env var
- SecurityDisabled Swagger docs (/docs, /redoc, /openapi.json) in production by default
- SecurityGated stress test endpoint behind X-Stress-Test header with mandatory STRESS_TEST_SECRET env var
- SecurityRemoved EXOGRAM_TEST_TOKEN bypass from rate limiter
- SecuritySanitized RFC-0001 and protocol pages — all references reflect deterministic Python logic gates
- AddedKeyword guide for ChatGPT page — trigger phrases for storing, recalling, and checking facts
- AddedKeyword guide for MCP page — same tutorial format with Claude Desktop instructions
- AddedCopy-to-clipboard prompts on ChatGPT and MCP pages (no more opening new sessions)
- AddedManual MCP setup instructions with JSON config, macOS/Windows config paths
- Added'Keep using same conversation' guidance for ChatGPT users
- ImprovedEnhanced vector context — top-5 related ledger records from Pinecone feed inference for deeper cross-referencing
- ImprovedInference quality scales with vault size — more stored ledger records produce richer, more accurate responses
- FixedPinecone $eq null namespace error (EXOGRAM-2Q) — vault search no longer fails on empty filters
- FixedFree tier billing badge now shows 'Active' instead of 'Inactive'
- FixedResolved libstdc++.so.6 build error for Cohere import (EXOGRAM-2N)
v2.1.2
February 22, 2026- AddedAI plugin manifest at /.well-known/ai-plugin.json
- AddedMonthly API usage reset cron endpoint
- AddedGDPR data export and account deletion endpoints
- FixedChrome extension now uses production API domain
- ImprovedAll V2 routes (billing, subscription, API keys) mounted in production
v2.1.0
V2 LaunchFebruary 2026- AddedV2 Control Stack — 4-layer architecture (ledger, Meaning, Judgment, Action)
- AddedDeterministic Judgment Engine with 7 gates
- AddedEscalation routing for blocked/escalated actions
- AddedCircuit breaker pattern for fault isolation
- AddedEntity extraction engine (EMAIL, PHONE, URL, DATE)
- AddedPII scrubbing before ledger writes
- AddedState hash computation for tamper detection
- AddedOpenAPI spec served at /openapi.json
- Improved32 red-team adversarial tests passing
v2.0.0
FoundationJanuary 2026- AddedSemantic ledger with immutable versioning
- AddedConflict detection across claims
- AddedPinecone vector search for semantic recall
- AddedModel-agnostic inference routing
- AddedChrome extension for AI assistant integration
- AddedMCP server for Claude Desktop integration
- AddedStripe subscription billing