Cortex v5.0.0 "Eureka" 🧠⚡

Eureka — the moment of breakthrough discovery. v5.0 transforms Cortex from a brain-only assistant into a full-fledged developer companion: it can now be extended, scheduled, backed up, and integrated with the tools you already use. Inspired by Claude Code's extensibility model.

📦 Download

First-time launch on macOS: right-click Cortex.app → Open (the app is not yet code-signed with an Apple Developer ID).

✨ Highlights

🔌 Extensibility Layer (Eureka core)

• MCP server — Cortex acts as a Model Context Protocol server. Other tools (Claude Desktop, OpenCode, etc.) can connect and use your local brain + skills as MCP tools.
• Artifact viewer with sandboxed React + HTML + SVG runtime, rendered inline in chat.
• Fork button — branch any conversation message into a parallel exploration thread.
• Global ErrorBoundary for graceful renderer crash recovery.

⏰ Scheduler & Background Jobs

• Cron-style scheduled tasks: embedding-freshness checks, sync jobs, autoscan retries.
• ScheduledTasksPanel — view, pause, resume, trigger jobs manually from Settings.
• Per-job structured logs persisted in SQLite for replay + debugging.

💾 Backup & Restore

• One-click backup of brain state — SQLite + Qdrant + settings packed into a single .tar.gz.
• Restore any backup file in seconds with full state recovery.
• Forward-compatible format — backups from v5.0.0 restore in future v5.x.

🔗 New Integrations

• Slack — connect a workspace; agents can read channels and send messages.
• Application Insights — query Azure telemetry/logs directly from chat.
• Software bundle import/export — share complete projects (code + brain + settings) as a single artifact.
• GitHub output router — smart routing of agent output to PR comments / issue threads.

🎨 Chat & UI Overhaul

• Chat store refactor — 1100-line monolith split into modular slices (conversation, message, thinking, mappers).
• TrainingIntelligencePanel mounted globally for cross-project visibility.
• ThinkingProcess redesign — streaming reasoning view with collapsible steps.

🛠️ Build, Performance & Fixes

• New dist:win and dist:all scripts — first-class Windows installer + multi-platform releases.
• Eliminated memory leaks in long-running chat sessions — re-render storms fixed.
• Lone-surrogate sanitization in LLM context — no more emoji-pair crashes.
• Smarter intent classifier — sharper chat / code-edit / search / plan detection.
• MCP manager — cleaner lifecycle + auto-reconnect on disconnect.

📋 Full Changelog

See CHANGELOG.md for the complete list.

Diff from v4.5.0: v4.5.0...v5.0.0

v4.4.0 "Thalamus" — Full-Stack Activation

Thalamus — the brain's relay station, routing signals to exactly the right region for processing.

What Changed

Previously, ~80% of queries bypassed Cortex's agent/skill infrastructure and went straight to RAG→LLM. This release fixes that.

Phase 1 — Smart Pipeline Routing

determinePipelinePath() now uses routing category as primary signal:

Confidence thresholds lowered: skill_chain 0.7→0.5, orchestrate 0.8→0.6.

Phase 2 — Category → Skill Auto-Activation

Skill directives injected into system prompt automatically per category — no slash command needed:

• deep → code-analysis + react-agent (trace call chain, root cause first)
• ultrabrain → plan-execute + react-agent (plan before implement)
• visual-engineering → playwright-browser
• writing → react-agent (structure, prose quality)

Phase 3 — Proactive Background Agents

• deep / ultrabrain → explore + librarian fire in parallel
• visual-engineering / unspecified-high → explore fires
• Results merge into context if ready within 2s

Phase 4 — Loop Auto-Activation

• Keywords "liên tục", "không dừng", "until done", "autonomous" → Ralph loop auto-starts
• Category ultrabrain + query > 300 chars → Ultrawork loop
• Completion detection from response content (hoàn thành / done / complete)

Phase 5 — Training Feedback Loop Closed

• notifyChatStarted() at pipeline entry
• notifyPostChat() + notifyChatEnded() after every successful response
• AutoScan scheduler now receives signal after every interaction

Model Routing Fixes

• routeToModel() cross-references live model status — never routes to quota_exhausted model
• Eliminated hardcoded gpt-4o-mini fallback causing 400 errors
• model-fallback hook uses dynamic ready models sorted by tier

OMO Infrastructure Integration

• Hook system fully wired: on:session:start, on:tool:call, session lifecycle hooks
• Plugin config (cortex-config.jsonc): per-agent/category model overrides, hook disable list
• 60s stale task detection + cleanup interval

Impact

Cortex v4.2.0 — Synapse

Synapse — the junction between neurons where signals are transmitted. This release connects Cortex to a broader universe of information: any document type, structured data, and the proven wisdom of the world's best AI systems.

What's New

Document Intelligence — inspired by microsoft/markitdown ⭐ 92K

Cortex can now read, index, and answer questions about documents — not just code.

New AI tool: cortex_read_document — agents can now call this to read any document at query time.

New cortex_read_document trigger — SUPERPOWERS_CORE now automatically suggests document reading when relevant.

Document indexing — PDFs, DOCX, and spreadsheets are indexed into Cortex's brain during project sync. Ask questions about your design docs, specs, and reports.

UI metadata badge — When an agent reads a document, the chat shows a styled header with file type icon, name, and metadata (page count, sheet names, title, author).

Agent Brain Upgrade — inspired by x1xhlol/system-prompts-and-models-of-ai-tools (37+ AI tools)

All 15 specialized agents now share 10 production-proven policies distilled from Claude Code, Cursor v2, Devin AI, and Windsurf Cascade:

Fixed

• CSV parser now RFC 4180 compliant — handles quoted fields with embedded commas
• PDF converter API corrected for pdf-parse v2.x
• XLSX converter uses RFC 4180 compliant parser for cells containing commas

Build from Source

git clone https://github.com/hoainho/cortex.git
cd cortex && npm install && npm run dist:mac

See CHANGELOG.md for full history.

What's New in v4.1.0

Filesystem Tools — 4 → 9 tools

• cortex_read_files — Batch parallel read via Promise.all, up to 10 files simultaneously
• cortex_grep_search — Regex search across the entire project with file paths and line numbers; auto-skips node_modules, dist, and binary files
• cortex_edit_files — Apply multiple edits across multiple files in a single call; reports N/M succeeded
• cortex_move_file + cortex_delete_file — Safe file management with sandbox enforcement
• cortex_read_file — File size limit raised from 1MB → 10MB; new offset+limit parameters for chunk reading large files
• cortex_list_directory — New recursive, depth (1–10), and extensions filter parameters
• Progressive edit fallback in cortex_edit_file: exact match → whitespace-normalized match → contextual line hints with actionable suggestions

Unrestricted Mode

• New setting: filesystem_unrestricted_mode (default: false)
• When enabled: AI can read/write any file on the machine, including absolute paths outside the project
• Still blocks: /System, /etc, /bin, /sbin, /var, node_modules
• Inspired by Cline/Roo Code architecture — warn instead of block

Agent Pool — 429 Rate Limit Resilience

• Dynamic model resolution: pool now queries getAvailableModels() at runtime instead of using hardcoded model names
• Tier-based selection: fast (tier 1–5), balanced (tier 5–7), premium (tier 8–10); picks the best available model in range
• Exponential backoff with jitter: 429/5xx errors retry after ~1s / ~2s / ~4s (±30% jitter)
• Retry-After header: respects server-provided delay when present
• Tier fallback: RESOURCE_EXHAUSTED in response body triggers automatic cascade fast → balanced → premium
• Auto-refresh: model list is refreshed before pool execution if no models are ready

Bug Fixes

• Planning skill response truncated at 2,000 chars → raised to 8,000 chars
• Agent Pool returning 0/2 succeeded on first 429 hit — now retries with backoff before failing

Build from source:

git clone https://github.com/hoainho/cortex.git
cd cortex && npm install && npm run dist:mac

See CHANGELOG.md for the full change history.

Cortex v3.0.0 — Agent Mode Overhaul + V3 Engine

Agent Mode

• Agent system prompts hidden from chat UI
• OpenCode-style mode directives: analyze-mode, search-mode, todo-continuation
• 4 named agents: Sisyphus, Hephaestus, Prometheus, Atlas

Smart Routing

• GitLab-first model priority (T10) with auto-rotation
• Model cache invalidation on proxy config change

V3 Hook Engine

• Before/after chat hooks: cost-guard, cache-check, context-window-monitor
• Loop engine with boulder state persistence

Download

• Cortex-3.0.0-arm64.dmg — macOS installer (Apple Silicon)
• Cortex-3.0.0-arm64-mac.zip — macOS portable (Apple Silicon)

Cortex v2.0.0

The AI Brain That Knows Your Codebase.

Features

• 12 specialized AI agents (Sisyphus, Hephaestus, Prometheus, Atlas + 8 core/background agents)
• 33 slash commands for workflows and skills
• Deep PR code review with 4 perspectives
• Knowledge crystallization and nano-brain memory
• Multi-model support with auto-rotation
• Jira, Confluence, and GitHub context integration
• Product Manager & Engineer dual modes

Download

• macOS: Download Cortex-2.0.0.dmg below

Installation

1. Download the .dmg file
2. Open and drag Cortex to Applications
3. Launch Cortex and import your first repo