Releases

• Autonomous Web Search: Enables the AI agent to search the web in real time for coding solutions, up-to-date documentation, and error fixes.
• Remote Web Ingestion: Fetches remote web pages and documentation directly into context as clean Markdown with zero tracking or dependencies.
• Live Web Research Integration: Empowers reasoning models to autonomously consult the live web during planning and execution.
• Safe Concurrent Tool Calling: Dispatches web searches and content fetches alongside workspace tools in parallel batches.
• Real-Time Streaming Support: Extracts and runs web research tool invocations directly from live model response streams.

• Comprehensive Workspace Tool Suite: Equips models with full autonomous capabilities to create, edit, inspect, and organize project files.
• ReAct Reasoning Protocol: Operates on an iterative Think, Act, Observe, and Reflect cycle with automated task classification.
• Automated Self-Correction & Verification: Automatically detects diagnostic errors after code edits and executes targeted fix passes.
• RAG Memory & Context Engine: Persists long-term project knowledge and dependency awareness locally in your workspace.
• Human-in-the-Loop Safety Gates: Provides configurable permission levels, diff previews, and approval prompts for destructive actions.
• Multi-Agent Orchestration & QA: Coordinates specialized Architect, Coder, Reviewer, and Security Auditor roles with automated quality scoring.
• Search-and-Replace Block Diffs: Emits targeted edit blocks to dramatically reduce output token consumption and speed up generation.
• Parallel Multi-Tool Calling: Dispatches multiple read operations simultaneously in a single turn to minimize round-trip latency.
• AST Skeleton Context Ingestion: Extracts interface contracts and type signatures to reference large files with minimal token overhead.
• Automatic Provider & Key Failover: Seamlessly reroutes requests across fallback API keys and providers when encountering rate limits or quota depletion.
• Tiered Model Cascading: Dynamically routes specialized sub-tasks to the most cost-effective model tier to preserve API budget.
• Progressive Tool Execution & Step-Budgeting: Enforces strict execution bounds and loop detection to prevent runaway agent turns.
• Project Rule Ingestion: Automatically discovers and applies repository instruction files to ensure generated code follows project conventions.

Technical Improvements & Security

• Enforces execution safety bounds, error classification, and intelligent output truncation on terminal commands.
• Automatically profiles workspace tech stacks, directory trees, and version control status on conversation start.
• Blocks hazardous shell commands and unauthorized path escapes outside your project workspace.
• Adds response feedback controls that feed developer evaluation preferences directly into workspace memory.

• Serverless Architecture: Communicates directly with official provider APIs with zero intermediate telemetry servers and complete user privacy.
• Sidebar Chat Interface: Provides an integrated IDE panel for multi-turn conversations, syntax-highlighted code blocks, and active file context attachments.
• Context Menu Actions: Enables instant code explanation, refactoring, and bug fixing directly from editor right-click menus.
• Core File Operations: Equips the assistant with essential tools to view, create, edit, search, and organize workspace files.
• Smart Context Awareness: Automatically incorporates active document contents and highlighted editor selections into AI prompts.
• Configurable Settings: Grants full control over default models, temperature, max token limits, and custom gateway endpoints.
• Extensive Model Support: Connects to 17+ leading AI providers including Google Gemini, OpenAI, Claude, DeepSeek, and Groq using personal API keys.