Overview

This page tracks every major feature shipped in Contox, organized chronologically by phase. Each feature links to its documentation page where available.

---

Phase 1: Core Platform

Status: Complete

The foundation — project management, client packages, and the API layer.

Platform

• Project management — Create and configure projects, manage API keys and team access
• Context system — Hierarchical contexts with full CRUD, search, and parent/child relationships
• REST API — Full API with HMAC signing, rate limits, and webhooks
• Authentication — Email/password auth with email verification and password reset

Client packages

• MCP Server — Model Context Protocol server for Claude Code and compatible tools (23 tools)
• CLI — Universal command-line interface for all AI tools (17 commands)
• VS Code Extension — Integrated experience with sidebar, context tree, and session management

Integrations

• Claude Code — Native MCP integration
• Cursor — MCP or CLI integration
• Windsurf — MCP or CLI integration
• GitHub Copilot — CLI integration
• Other tools — Any MCP-compatible or context-file-based tool

Dashboard

• Team management — Invite members, assign roles, manage seats
• Billing — Stripe-powered plans, credits, invoices
• Admin panel — Admin-only project and user management
• Documentation site — 125+ pages with Cmd+K search

---

Phase 2: V2 Memory System

Status: Complete

The enrichment pipeline that turns raw session data into structured, searchable knowledge.

Capture pipeline

• Session capture — VS Code extension watches git commits and file saves, sends events with diffs to the ingest API
• Git digest — SHA-based range tracking extracts commit messages, file changes, and smart patches
• Codebase scanner — contox scan analyzes your project structure and generates 15–20 structured sub-contexts

Enrichment

• AI-powered enrichment — Extracts structured knowledge from raw events: architecture, conventions, decisions, bugs, todos
• Quote verification — Programmatic check rejects memory items with hallucinated citations
• Plan-based model selection — Free/Personal plans use Small model, Team+ uses Medium for higher quality

Brain & retrieval

• Brain assembly — Tiered knowledge document with 3 layers (active, reference, archive)
• Schema key taxonomy — 30 canonical keys with LLM-based and deterministic classification
• Semantic search — Embedding-powered search across all memory items
• Context packs — Token-budgeted, semantically relevant context for AI sessions

Quality & maintenance

• Hygiene agent — Two-phase workflow: analyze then apply (rename, retag, merge, deprecate, redact)
• Memory items — Confidence scores, importance ratings, staleness tracking, archive/review states
• Evidence truncation — Optimized diff capture (3K chars/commit) and evidence limits (2K chars) for performance

---

Phase 3: Genesis & Dashboards

Status: Complete

Full codebase analysis, rich dashboard pages, and unified memory across all sources.

Genesis Scan

• Genesis Scan — 8-phase AI pipeline that analyzes your entire codebase: product identity, architecture, patterns, data flow, entry points, dependencies, security audit, and risk analysis
• Unified memory — Genesis findings merged into the same schema key taxonomy as session-enriched items — no more separate silos

Dashboard pages

• Findings Explorer — Browse all memory items across 27 category tabs with source badges (Scan/Session/Manual), freshness indicators, confidence filtering, category heatmap, and knowledge source distribution
• Security Center — Dedicated security view with vulnerability scoring (A–F grade), severity breakdown, issue categories, most affected files, and auto-dismiss via recent git commits
• Brain Explorer — Interactive force-directed knowledge graph, hierarchical tree view, category bubble chart, memory freshness audit, and inline item editing
• Ask (AI Chat) — Natural language Q&A over your project memory with cited sources and confidence scores
• Sessions — Session detail view with files modified, captured events, and timeline
• Operations — System health monitoring, flagged items, pipeline status

Intelligence features

• Security vs Risks split — Vulnerabilities (OWASP-style issues) separated from Risks (technical debt and improvement suggestions)
• Source tracking — Every finding shows its origin: Genesis Scan, session enrichment, or manual entry
• Freshness indicators — Relative timestamps on every item showing when it was last updated
• Category heatmap — Visual breakdown of knowledge concentration across your project

---

Phase 4: Quality & Scale

Status: Complete

Multi-pass dedup pipeline, analysis quality gates, onboarding, and developer experience improvements.

Quality pipeline

• Extended noise patterns — isNoiseFinding() filters generic titles, short descriptions, and "Uses X" patterns across all layers
• Layer-specific quality prompts — Each Genesis layer has aggressive exclusion rules: architecture rejects "Component-based", dependencies skips standard packages, entry points consolidate by API prefix
• Embedding dedup pass — Pass 4 uses Mistral Embed at 0.88 cosine similarity threshold with pre-clustering for large sets (1000+ findings)
• Post-assembly AI quality gate — Pass 5 sends all titles to Gemini to identify remaining noise and duplicates (runs when > 20 findings)
• Cross-source batch dedup — Slug-based and embedding-based matching merges overlapping genesis and session findings, superseding lower-quality duplicates

New pages

• Genesis Live View — Real-time scan progress with 8-phase stepper, file/chunk counters, duration tracking, and auto-polling
• Onboarding flow — 5-step guided setup: create organization, first project, run Genesis scan, connect IDE, ready to go
• Quality dashboard — Scan history with quality scores, findings trends over time, duration and cost tracking

Platform enhancements

• Compact brain summary — assembleCompactBrain() produces ~2-3K token summaries vs ~20K full brain, used by default in contox_get_memory
• Memory freshness system — staleness-checker.ts detects items with deleted files (stale) or heavily modified files (review), integrated with compaction pipeline
• Auto-context injection — VS Code extension watches active files and pushes semantically relevant memory to AI tools via .contox/context.md and rule file injection (.cursorrules, .windsurfrules, etc.)

---

Up Next

Status: Planned

Advanced intelligence features and experience improvements.

• Architecture drift detection — Compare sessions against Genesis baseline, detect pattern violations and convention drift Done
• Smart onboarding guide — Auto-generated new developer guide with architecture overview, conventions, key files, and gotchas Done
• Public status page — Real-time service health monitoring for all Contox services (database, worker, integrations) Done
• Impact radar — Analyze change impact radius across modules, show which components are affected by a file modification Planned
• Smart changelog — Auto-generated changelog from sessions with different audience modes (engineer vs stakeholder) Planned
• Light mode for documentation Planned
• Mobile-responsive dashboard Planned

---

Feature summary

Next steps

• Changelog — Detailed version history and release notes
• Introduction — Get started with Contox
• Genesis Scan — Run a full codebase analysis