Changelog
Every release, every feature. From day one to now.
July 2026
v3.5.9
A Fresh New Look & Smarter Than Ever
- Brand-new interface — cleaner, faster, and built for focus
- Settings redesigned from the ground up: everything you need, exactly where you expect it
- Universal MCP support — connect with VS Code, Cursor, Windsurf, Kilo, and any MCP-ready tool
- Blazing fast sessions and buttery-smooth rendering — the snappiest Codey yet
- Full reasoning control — pick your effort level from none all the way to max
- Bring your own provider with Anthropic or OpenAI compatibility modes
- Smarter thinking across the board — optimized for Alibaba, MiniMax, Anthropic, Hugging Face, Z.ai, and more
July 2026
v3.5.8
Bug Fixes & Stability Improvements
- Resolved various minor bugs and improved overall application stability.
- Addressed internal styling and layout issues for a smoother user experience.
June 2026
v3.5.7
Smooth Collapsible Work Logs & Live Timer
- Introduced a collapsible "Worked" container for intermediate actions (tool calls, reasoning steps) to reduce chat clutter.
- Added a live, ticking elapsed-time counter in the header that accurately tracks duration while the AI is executing tasks.
- Implemented smooth, spring-driven expand and collapse transitions with custom easing (SOFT_SPRING) and subtle organic bounce.
- Enhanced motion design to animate in all scenarios, including manual toggle of historical or finished session turns.
- Resolved layout jumps and visual snaps by moving spacing to the inner padding of the transition container.
June 2026
v3.5.6
Agent Memory & Open Session
- Durable memory system (SessionMemory) — AI retains context and facts across sessions and subagents
- New Memory tool with recall, remember, replace, forget, list actions across 3 scopes: user, project, agent
- Auto-extract: AI automatically learns and stores durable facts after each step — no manual saving needed
- Unified memory protocol across all agents (Prometheus, Athena, Matis, Hermes) with clear read/store/forget rules
- Memory config: max 25KB storage, 120 entries limit, auto-cleanup after 120 days, per-agent permissions
- New Open Session tool — spawn a child session with any agent and model (GPT-4o, Claude Sonnet, Gemini, DeepSeek, and more)
- Interactive Open Session UI in both TUI and App — agent name, model, thinking status, and navigation to child session
- New IRIS and SCOUT subagents with advanced task delegation system
- Prometheus promoted to default implementation agent
- Major agent improvements: Prometheus (implementation), Athena (planning), Matis (research), Hermes (workflows)
- Granular memory permissions — configure read/write access per agent
- Faster agent response times and reduced token consumption
- Memory context handoff in task delegation — subagents inherit relevant durable facts automatically
- Memory auto-extract uses structured AI generation (generateObject) for reliable, deduplicated fact extraction
June 2026
v3.5.5
Automation Tasks
- AI agents can now create, manage, and control automation tasks end-to-end
- All agents (Prometheus, Athena, Matis, Hermes) gain automation awareness
- Agents detect workflow patterns and suggest n8n automation when appropriate
- Talos subagent handles full n8n workflow lifecycle — from design to deployment
- 20 n8n tools available including workflow creation, validation, and execution
- Dual-mode system: documentation mode (7 tools, no credentials) and full mode (20 tools)
- 2,700+ n8n templates searchable for instant workflow scaffolding
- Multi-service integration: connect Slack, GitHub, Mailchimp, Notion, and more
- Scheduled workflows, event-driven triggers, and data transformation pipelines
- Production-grade validation with 4 profiles (minimal, runtime, ai-friendly, strict)
- Automatic error handling injected on all external API calls
- Workflow lifecycle management: create, validate, deploy, monitor
May 2026
v3.5.0
Built-in Tools & Integrations
- Playwright browser automation integrated as always-available MCP server
- 20+ browser tools: navigate, click, type, drag, screenshot, form filling
- Filesystem MCP server with advanced file operations and batch processing
- Parallel multi-file reading, directory trees, glob-based search
- Safe editing with dry-run preview mode before applying changes
- Hermes agent (Workpilot) becomes primary consumer of Filesystem MCP
- Browser tab management, keyboard input, and file upload support
- Page analysis tools: snapshot, accessibility tree, console logs, network requests
- Auto-initialization of built-in MCP servers on startup — zero configuration
April 2026
v3.4.0
Workpilot — Hermes Agent
- Introducing Workpilot: a dedicated mode for non-code work
- Hermes agent handles research, documents, spreadsheets, slides, PDFs, and browsing
- PDF reading, extraction, and generation
- Excel/CSV creation and manipulation with formula support
- Document drafting and editing (DOCX, Markdown)
- Presentation and slide deck generation
- Browser-based research and web navigation
- n8n workflow automation from within the workspace
- Filesystem MCP integration for advanced multi-file operations
- Drag-and-drop, form filling, and keyboard automation via Playwright
March 2026
v3.3.0
Autopilot — Matis Agent
- Introducing Autopilot: end-to-end application generation from prompt to production
- Matis agent takes an app idea and builds a complete Next.js product
- Infinite canvas architecture connecting PRD, database, UI, components, APIs, and deployment
- Production-ready web surfaces with premium UI direction built in
- Full-stack integration: database schemas, webhook endpoints, API responses
- Automatic n8n workflow planning for app features
- Design taste embedded in the generation flow — no generic templates
- Delegation protocol between Matis and Talos for automation components
February 2026
v3.2.0
Prompt Engineering & Agent Intelligence
- Complete rewrite of all agent system prompts for concision and role fidelity
- Prometheus: enhanced build-mode with n8n workflow detection patterns
- Athena: strict plan-only role with anti-manipulation boundaries
- Scout: read-only exploration with configurable thoroughness levels
- Iris: specialist agent for background parallel work
- Cross-agent integration of simplicity principles and coding standards
- Improved tool documentation consistency across all prompts
- Behavioral constraints embedded into agent prompts for safety
- Language matching: agents respond in the same language as the user
- Tool playbook system for structured tool usage guidance
January 2026
v3.1.0
Performance & Stability
- Smooth scroll acceleration in Terminal UI (Mac-style gesture response)
- Animation refinements across desktop and web interfaces
- Improved error recovery across all agent workflows
- Session stability improvements for long-running conversations
- Faster model provider initialization and SDK caching
- File read caching with mtime validation for performance
- Memory usage optimization for large project contexts
- Better handling of provider rate limits with graceful fallbacks
- UI polish: improved spacing, contrast, and micro-interactions
- Bug fixes for git snapshot operations and session forking
December 2025
v3.0.0
Multi-Agent Architecture
- Introducing the multi-agent system with 7 specialized AI agents
- Prometheus: the Builder agent for code implementation and refactoring
- Athena: the Strategist agent for architecture planning and analysis
- Scout: the Explorer agent for read-only codebase investigation
- Iris: the Specialist agent for background parallel tasks
- Matis: the Creator agent for full-stack application generation
- Hermes: the Operator agent for documents, research, and automation
- Talos: the Automation subagent for n8n workflow architecture
- Subagent system for context isolation and parallel execution
- Agent delegation protocol for cross-agent task routing
- Read-only vs build-mode agent separation for safety
November 2025
v2.5.0
Skills System
- Modular skill system for specialized domain expertise
- File-based skills defined in SKILL.md with YAML frontmatter
- Multi-source discovery: project, global, Claude-compatible, and custom paths
- 10-50x faster search with inverted indexes (5-10ms for 1000 skills)
- AI-powered contextual skill suggestions based on user prompts
- Automatic trigger matching — skills activate when relevant
- Bundled resources: scripts, references, and assets per skill
- Permission integration for skill access control
- Background sync with stale-while-revalidate pattern
- @mention and /skill slash command for manual skill loading
October 2025
v2.4.0
MCP Protocol Support
- Full Model Context Protocol (MCP) implementation for external tool servers
- Local MCP servers via stdio transport — run tools as subprocesses
- Remote MCP servers via HTTP/SSE with streaming support
- Built-in OAuth 2.0 authentication flow with PKCE
- Dynamic client registration and automatic token refresh
- Runtime tool discovery from connected MCP servers
- Resource and prompt support alongside tools
- Tool caching for performance optimization
- Tier-based access control for built-in MCP servers
- Configuration via codey.json with hot-reload support
September 2025
v2.3.0
Custom Agents & Subagents
- Define custom agent personas using markdown files
- Global agents in ~/.config/codey/agents/ or per-project in .codey/agents/
- Agent configuration: custom prompts, model selection, tool scopes
- Built-in lock guard: native agents (Prometheus, Athena) cannot be overwritten
- Global agents CRUD API for programmatic management
- Subagent spawning for parallel isolated tasks
- Subagent context isolation — agents cannot read each other's conversations
- Agent mode filtering: primary agents vs subagents vs hidden
- Migration-safe loading of legacy agent configurations
August 2025
v2.2.0
Session Intelligence
- Session compaction: automatic summarization of old messages to preserve context
- Session forking: create child sessions from any conversation point
- Session reverting: roll back to previous states with git snapshot integration
- Session sharing: generate shareable links to conversations
- Persistent conversation history across CLI and Desktop sessions
- Message-part architecture: text, tool states, files, reasoning, subtasks
- Real-time streaming of all LLM responses with token-level updates
- Tool execution logging with duration, input arguments, and output
- Thread history search and resume across sessions
- Visual diff reviews for file changes proposed by build agents
July 2025
v2.1.0
70+ AI Providers
- Unified interface for 30+ LLM providers via Vercel AI SDK and Models.dev
- Support for OpenAI, Anthropic, Google Gemini, Amazon Bedrock, and more
- OpenRouter integration for access to hundreds of models
- Ollama and local model support for fully offline operation
- Dynamic model discovery from provider APIs
- Adaptive thinking support for new Claude reasoning modes
- Provider-specific schema transformations for optimal compatibility
- API key, OAuth, and AWS IAM credential management
- Per-provider configuration overrides (timeouts, base URLs, custom headers)
- Model capability detection: text, images, audio, video, PDF input support
- Streaming reliability with SSE wrapping for consistent token delivery
June 2025
v2.0.0
Security & Permissions System
- Three-tier permission system: allow, ask, deny
- Granular control per tool, per directory, per operation
- Glob/wildcard pattern matching for flexible rules
- Protected directory enforcement — sensitive paths are read-only
- Bash command analysis: dangerous patterns (rm -rf) blocked before execution
- External directory protection with explicit approval required
- Session-based permission caching to reduce approval prompts
- Configuration via codey.json with project-level overrides
- Guard utility wraps all tool operations with automatic permission checking
- Structured error responses with permission details and matched rules
May 2025
v1.5.0
Desktop Application
- Native desktop application built with Tauri for macOS, Windows, and Linux
- Bidirectional sync between Desktop and CLI sessions
- Real-time collaboration through git worktree management
- Dual-authentication: provider keys (auth.json) + Codey account (account.json)
- Subscription tier management synced between Desktop and CLI
- Cross-platform installation: Homebrew, npm, Scoop, Chocolatey, and curl
- Auto-update system with version management
- System tray integration and native notifications
- Hardware-accelerated rendering for smooth UI performance
April 2025
v1.4.0
Web Interface & Dashboard
- Rich web client dashboard accessible via local server
- Thread history navigation with search and resume
- Visual diff reviews with color-coded file changes
- Tool execution logs: click any step to review output and duration
- Live settings management for models, themes, and experimental flags
- Real-time streaming of agent responses in the browser
- WebSocket-based event system for live updates
- Responsive layout for desktop and tablet browsers
March 2025
v1.3.0
LSP & Code Formatters
- Language Server Protocol integration for syntax and compilation analysis
- Automatic diagnostic feedback: agents detect and fix compile errors
- Support for TypeScript, Python, Go, Rust, and more via LSP servers
- Auto-triggered code formatters after file edits (Prettier, ESLint)
- Custom formatter commands configurable per project
- Formatter detection from project configuration files
- Tree-sitter parsers for syntax highlighting in terminal UI
- Code search powered by ripgrep for fast text matching
February 2025
v1.2.0
Terminal UI (TUI)
- Interactive Terminal UI with streaming responses and visual file previews
- Color-harmonized themes: codey, one-dark, everforest, synthwave84, and more
- Customizable keybindings with leader key system (Ctrl+x default)
- File reference system: type @filename to inject file contents into prompts
- Shell command integration: prefix with ! to feed output directly to agents
- Keyboard shortcuts: new session, undo/redo, compact, editor, share, exit
- Agent switching via Tab key between Build and Plan modes
- Scroll acceleration with Mac-style smooth gestures
- Fuzzy search auto-completion for slash commands
January 2025
v1.1.0
Core Tool System
- 9 file editing strategies for surgical code modifications
- Shell execution with infinite loop detection and background process support
- File reading with automatic truncation for LLM context limits
- File writing with atomic operations and backup
- Glob-based file discovery and ripgrep-powered text search
- Web fetching and web search tools for real-time information
- Code search for semantic code navigation
- Todo management for task tracking within sessions
- Truncation-aware output handling for large results
- All tools validated with Zod schemas and permission-checked before execution
December 2024
v1.0.0
Initial Release
- Local-first AI coding agent for terminal and desktop
- Project-aware context with automatic git integration
- Configuration hierarchy: remote, global, custom path, project, inline, managed
- Support for JSON and JSONC configuration files
- File-based JSON storage engine with file locking for concurrency
- Event bus system for decoupled communication between modules
- Instance-based architecture with state isolation per directory
- Git worktree management for parallel session support
- Structured logging with service-based organization
- Feature flags for gradual feature rollout
- ID generation, shell detection, and environment utilities
- Cross-platform support: macOS, Linux, and Windows
Initial release — December 2024