Looking for Zenflow Desktop updates? Check out the Zenflow Changelog.
Faster Agents, Tighter Permissions, and a Cleaner Plugin Experience
Auto-Approve Mode for Agent Actions
A new Auto-approve mode is now generally available across VS Code and JetBrains. Ideal for trusted workflows where you want the agent to keep moving without manual confirmation at each step.Permission Management for MCP Tools
Agents now request permission before using MCP tools. Grant one-time access or configure “Always allow” for trusted tools in settings — full visibility into which external tools run during a task.Hooks in ZenCLI
A new hooks system inzencli lets you run custom commands at key points in an agent’s lifecycle — perfect for logging, custom telemetry, or pre/post-task automation.Global Installation of Codex and Claude Code
The CLI now supports global installation of Codex and Claude Code, making it easier to share runtimes across projects without re-downloading on every run.Provider API Keys from .env
Configure provider API keys directly in your project’s .env file — the CLI will pick them up automatically.Improvements
- Custom rules removed — use
AGENTS.mdinstead for a single, standardized way to define project conventions - Built-in system agents removed from VS Code and JetBrains — all tasks now run through the unified Coding Agent
- Usage-based billing support in plugins
- More consistent sign-in experience in VS Code across all auth flows
- Fixed JetBrains compatibility issues affecting newer IDE versions
- Fixed plugin window constantly stealing focus during agent runs
- Fixed CLI hanging after task completion
New Models, Smarter Agents, and Better Control
GPT 5.5 Support
The latest frontier model from OpenAI is now available in Zencoder. Use GPT 5.5 for your most complex tasks — planning, debugging, and research workflows.Claude Opus 4.7 with Streaming Thinking
Claude Opus 4.7 is now supported with streaming thinking — watch the model’s reasoning process in real time as it works through your problem.Auto Permission Mode
A new Auto permission mode for Claude Code lets agents execute without manual approval at each step. Ideal for trusted workflows where you want the agent to move fast.Image Generation
Image generation is now enabled by default. Ask the agent to create visuals — logos, diagrams, mockups — and see results rendered inline in chat.Agent-Browser Skill
A new agent-browser skill replaces the previous Playwright skill, giving agents improved browser automation capabilities for testing and web interaction tasks.Frontend Design Skill
A new built-in frontend-design skill helps agents generate production-quality UI designs and HTML mockups directly from your descriptions.MCP Tool Permissions
Agents now request permission before using MCP tools, giving you visibility and control over which external tools are invoked during a task. You can also configure “Always allow” for trusted tools in settings.Delete Rules from UI
You can now delete skills directly from the VS Code and JetBrains UI — no need to manually remove files.Improvements
- Improved model selector — clearer display of the current engine and model
- Optimized system prompts for Claude 4.6 and GPT 5.4 model families
- Updated comprehensive review and cross-review skills
- Fixed agent reporting it edited a file when it didn’t actually make changes
- Fixed Anthropic models stopping prematurely due to output token limits
- Fixed long-running agent sessions failing with MCP tools
- Fixed empty
.zencoder/workflowsfolders being created in every project - Fixed session data corruption in VS Code
Smarter IDE Workflows and Collaboration
Multi-Model Code Review
You can now get a second opinion on your code using another model directly inside your IDE. Review with another model lets you validate and improve generated code quality. /comprehensive-review provides deeper, multi-model feedback for more critical changes.Skills in the IDE
Add and manage skills directly from the UI. Built-in skills are easier to discover and use, enabling more structured and repeatable workflows.Project Understanding with /init
A new /init command helps the agent understand your repository faster. Automatically gathers project context and structure, improving response quality from the very first interaction.Visual Code Understanding (Mermaid Support)
Agents can now render Mermaid diagrams directly in your IDE — useful for understanding architecture, flows, and generated outputs.IDE Notifications
Sound and visual notifications when responses are ready or approvals are required. Fully configurable in settings.Improvements
- More reliable task execution with improved handling of complex workflows
- Better subagent coordination for more accurate and structured outputs
- Smoother chat experience with improved auto-scrolling
- Simplified controls for model and tool selection
- Improved responsiveness across both VS Code and JetBrains plugins
Bug Fixes
- Fixed rare cases where the agent could freeze or crash
- Fixed issues with chat auto-scrolling and output rendering
- Fixed issues affecting task execution consistency
Important Changes
- Custom rules are being phased out in favor of a more structured approach
- Traditional inline code completion is no longer supported — focus is now on agent-driven workflows
February 2026
Gemini 3.1, GPT-5.3 Codex, Sonnet 4.6, built-in skills, research agent, subagent streaming
IDE Agent
New Model Support
Three powerful new models: Gemini 3.1 with improved thinking defaults, GPT-5.3 Codex for enhanced code generation, and Sonnet 4.6 for Anthropic users. The agent can now intelligently choose which model to use for subtasks.Built-in Skills
Three skills included by default: Playwright for browser automation, Code Review for structured review workflows, and Skill Creator for creating custom skills. Access via/ and @-menu.Research Agent
A dedicated research agent for exploration and analysis — codebase investigation, issue analysis, and pre-implementation research.Subagent Streaming
See subagent work in real time with live streaming output, clear task delegation visibility, and proper permission handling.Improvements
- Redesigned navbar with “All Chats” moved to the left
- Clickable chat title for quick rename/navigation
- Skills menu showing configured skills
- AGENTS.md support for agent instructions
- Review with another model for second opinions
- Adaptive thinking support for Anthropic provider
- MCP caching and lazy loading for faster startup
- Improved CLI startup time, migrated CLI to Bun
- Trial quota exhausted notifications
Bug Fixes
- Fixed “Wrapping up” hang and agent startup hangs
- Fixed incomplete tool call handling
- Fixed MCP error handling preventing deadlocks
- Fixed revert button visibility in JetBrains
- Fixed API key inheritance billing issues
- Fixed token visibility in IDE logs
Skills Support
Create reusable prompt templates that modify how the agent responds. Local and global skills in.agents/skills (project) or ~/.agents/skills (global).Terminal Mentions
Reference terminal content directly in chat — share command results, error logs, and build output with the agent.Chat Navigation Dropdown
Quick chat switching, search functionality, and pin important chats for easy access.Context Window Indicator
Real-time percentage display showing how much of the context window is being used. Helps optimize prompts and avoid context limits.Improvements
- Click to open files mentioned in chat responses
- Streaming indicators and action required indicators
- Improved small screen UI and better error recovery
Bug Fixes
- Fixed chat deletion redirect, agent selection on edit
- Fixed mention popup cleanup, workflow card spacing
- Fixed table line breaks, chat input focus
- Windows compatibility fix for some users
Spec-Driven Development Workflows
Custom workflows for spec-driven development — guided development process, spec-first approach, and improved reliability through systematic planning.Enhanced Markdown Support
Markdown tables rendering, copy as Markdown button, enhanced code blocks, and line break preservation.Model Context Protocol Enhancements
Per-project MCP configuration, better session management, enhanced reliability, and RAG search integration.Platform Stability Improvements
Faster agent responses, better edge case handling, improved memory usage, and enhanced long-session reliability.Custom Models in Zencoder Plugins
Use privately deployed models or connect to unsupported providers. Full flexibility to leverage custom LLMs while maintaining Zencoder’s development features.Lines of Code Metrics
New LoC metrics in the usage analytics dashboard — lines of code generated vs. lines of code accepted. Available in the web admin panel and Analytics API.Dynamic Model Switching
Change AI models between messages without losing context or starting a new chat session.Model Catalog Updates
New: GPT-5.1-Codex, Gemini Pro 3.0. Removed: Sonnet 4 PT, GPT-5 (superseded by GPT-5.1-Codex).Clipboard Image and Text Pasting
Paste content directly from clipboard into chat — images, screenshots, long text, and code blocks. Cross-platform support across VS Code, JetBrains, and webview.Additional Updates
- Code blocks no longer collapse during streaming
- Clickable mentions for quick navigation
- Code hunk language detection for syntax highlighting
- Removed deprecated docstrings feature
- MCP tool reliability improvements
- 20+ bug fixes across API handling, chat, model selection, and platform-specific issues
Universal AI Platform
Connect your existing ChatGPT, Claude, or Gemini subscriptions to access Zencoder’s enterprise features. Native CLI support with Claude Code and OpenAI Codex. Works in VS Code (2.38+) and JetBrains (2.16+).Sonnet 4.5 Parallel Thinking
Day-zero support for Anthropic’s latest model — parallel execution, persistent state tracking, high verification bias. Available on Starter+ plans at 1.5× multiplier.Cost-Efficient Model Options
Grok Code Fast 1 at 0.25× multiplier — the most cost-efficient model in the lineup. Plus GPT-5 Codex at 1× multiplier.Comprehensive Troubleshooting
New guides for VS Code Gray Screen (GSOD) and Windows CLI Installation issues.Analytics Dashboard
Organizational insights for Core plans and above — daily active users, IDE and language analytics, member activity tracking.Enhanced Model Selection
GPT-5 support added within 24 hours of release. Gemini 2.5 Pro at 0.75× multiplier. Opus 4.1 on Advanced/Max plans. BYOK support for OpenAI and Anthropic.Skills Enhancements
Custom skill folders for compatibility with Cursor, Windsurf, Cline, and Continue. Create skills from chat. Currently applied skills badge.MCP Improvements
Remote and HTTP-based MCP support with OAuth2 for JetBrains. Max plan with priority support and SLA guarantees.Multi-Repository Search
Index multiple repositories through the web admin panel. Cross-repository code understanding with automatic context detection.Project-Specific AI Instructions
Project-level instructions for team sharing. Conditional application with glob patterns (now evolved into Skills).Image Attachments for All Users
Universal access to image attachments across VS Code and JetBrains. Design-to-code workflow with repository image references.E2E Testing Enhancements
Multi-framework support — Cypress, Playwright, or Selenium.E2E Testing
Automated E2E test creation in chat. Playwright integration, intelligent test recording, cross-browser testing.New Pricing Model
LLM call-based pricing instead of token usage for better transparency and predictability.Project Context
Comprehensive project analysis for context. Reduces LLM calls and speeds up agent responses.Image Prompts in VS Code (Beta)
Repository image support for visual context — upload Figma screenshots or design mockups and generate matching components.Zen Agents Platform
Agent sharing across organizations, custom agent registry synced across devices/IDEs, and tool configuration for custom agents.MCP Library and Tools Management
Built-in library with 100+ MCP servers. Project-level MCP installation. Dedicated Agent Tools UI.Zen Agents Marketplace
Browse 20+ purpose-built agents. Community contributions via open-source repository. Available in IDEs and on the website.Change Management
New ability to revert and rollback agent changes with granular control.Bash/Shell Tool for Coding Agent
Bash tool now available to the coding agent with safe command UX, “Don’t ask again” option, and Coffee Mode support.Requirements Tool
Automatically asks for clarifications, delivers higher-quality results, and reduces back-and-forth.Enhanced Chat Experience
Slash commands (/) and file mentions (@) hints in chatbox. Warning system for long chats. Attach files from outside the project.Bug Fixes
- Fixed indexing stuck at “Updating index”
- Fixed cursor positioning above inserted code snippets
- Jira authentication flow improvements