Found a bug? Report it on GitHub

Release Notes

Changelog

EN ES

A record of all notable changes, improvements, and technical updates to TukiCode.

v1.3.3 Stable

The High-Performance Architecture Update

TukiCode v1.3.3 is a major architectural milestone introducing zero-cost Rust bindings for critical operations, parallel execution, and a completely redefined background job management system.

Rust Native Engine

  • tuki_native Integration: Replaced heavy Python OS operations with a custom Rust extension. File system walking, searching, and PTY I/O are now executed with zero-cost bindings, significantly reducing TTFT (Time To First Token) on large repositories.

Execution & Optimization

  • Parallel Tool Execution: The agent can now invoke and process multiple tools simultaneously in a single turn, drastically reducing latency in complex workflows.
  • Prompt Caching (Anthropic): Added native support for Anthropic context caching, slashing token costs for repeated queries and instantly speeding up follow-up interactions.
  • True OS-Agnostic Engine: The prompt generator dynamically detects the underlying OS (Windows, macOS, Linux) to tailor shell commands and paths natively.

Stability & UX

  • Timeout Recovery (Async Handoff): Long-running synchronous commands (like winget or heavy npm installs) that time out no longer cause the agent to fail. They gracefully transition into background jobs, returning the PID to the agent for continuous monitoring while the UI remains unblocked.
  • Anti-Loop Exemptions: Refactored the Anti-Loop mechanism to whitelist passive polling tools (e.g., get_process_output). The agent can now safely monitor background installations indefinitely without triggering false-positive loop penalties.
  • Autonomy Overhaul: Completely removed the deprecated risk_level parameter, transitioning the entire codebase to use the simpler and safer global autonomy_level.
v1.3.2 Stable

Linux Compatibility & Installation Improvements

TukiCode v1.3.2 addresses critical compatibility issues on Linux systems and enhances the installation experience across all platforms.

Bug Fixes

  • Fixed Linux Import Error: Resolved a ModuleNotFoundError where textual.widgets._tab was missing in the Linux binary.
  • Pinned Dependencies: Fixed the textual version to 8.2.5 to ensure stability and avoid breaking changes from internal framework updates.
  • PyInstaller Optimization: Updated the build configuration to explicitly collect all Textual submodules.

Installation & UX

  • Improved Success Message: The installation scripts (install.sh and install.ps1) now feature a cleaner, more professional success message with direct links to documentation.
v1.3.1 Stable

Improvements to Build, Plan and Chat Modes

TukiCode v1.3.1 introduces important fixes and behavior improvements across the different working modes.

Build Mode Improvements

  • Fixed the execution flow of Build Mode.
  • Build Mode can now automatically start executing and developing a generated plan without requiring user confirmation.
  • Improved separation between Build Mode and Plan Mode.
  • Build Mode is now focused on autonomous planning + execution workflows, while Plan Mode remains confirmation-based before execution.
  • Improved plan continuation and execution reliability during long-running tasks.

Plan Mode Fixes

  • Fixed issues related to plan generation and confirmation flow.
  • Improved planner stability and execution handoff.

Chat Mode Fixes

  • Fixed minor issues affecting direct conversational execution.
  • Improved runtime stability in standard chat interactions.
v1.3.0 Stable

The Asynchronous Engine Update

A complete architectural rewrite moving to a non-blocking asynchronous execution model, strict MVC separation, and highly reliable structured JSON outputs for planning.

Architecture & Stability

  • Fully Asynchronous LLMs: All AI providers (Anthropic, OpenRouter, Gemini, Ollama) now use native async execution (httpx, asyncio), dramatically improving responsiveness and streaming speeds.
  • MVC Separation: Business logic and AI orchestration have been decoupled from the TUI and moved into a centralized TukiController.
  • Chat, Plan & Build Modes: TukiCode now operates in three distinct modes. Chat Mode is a direct conversational interface for quick questions, code explanations, and instant edits. Plan Mode breaks complex tasks into atomic, sequential steps — the agent generates a full roadmap, asks for your confirmation, and executes each step independently, making large refactors and project migrations reliable and resumable. Build Mode gives the agent full autonomy to execute the approved plan end-to-end without interruptions.
  • Structured Outputs: The Planner module now enforces strict JSON format natively (response_format) to guarantee 100% reliable step generation.
  • OpenRouter Error Recovery: Automatic fallback and retry mechanisms to gracefully handle OpenRouter models that do not support strict JSON output flags.
v1.2.3 Stable

The Cross-Platform Update

The official release marking the transition from Early Access to a stable, multi-platform tool. This release brings full support for macOS and Linux alongside significant deployment improvements.

Cross-Platform Support

  • macOS & Linux Binaries: Full standalone executables are now built and distributed for Unix-like systems.
  • PTY Abstraction: Completely refactored the interactive terminal architecture to detect the OS natively, substituting pywinpty and cmd.exe with Unix ptyprocess and default shells on macOS/Linux.

Distribution & Website

  • OS-Aware Downloads: The new homepage dynamically detects your operating system and offers the right installer.
  • Universal Installation Scripts: A new install.sh securely installs the binary on Unix and adds it to your PATH automatically.
  • Public Release: Removed Early Access gating; TukiCode is now publicly available for everyone to install.
v1.2.0 Early Access

The Terminal Update

A major release focused on interface redesign, interactive terminal support, and an improved configuration experience. This is the first Early Access release of v1.2.0.

Interface

  • Full TUI Redesign: Rebuilt using Textual — now a fullscreen, panel-based interface replacing the previous Rich-based output.
  • Three-Column Layout: Chat panel (center), file explorer (left), and Live Console (right) running simultaneously.
  • Live Console Panel: Dedicated terminal output panel powered by a pyte VT100 emulator. Toggle with Ctrl+B.
  • Status Bar: Persistent bar showing active model, risk level, and token count at all times.

Terminal & Execution

  • PTY Support via pywinpty: Processes like npm start and expo now run inside a real pseudo-terminal.
  • VT100 Emulation: Raw PTY output — including cursor movements — is processed through a virtual screen buffer before rendering, enabling correct QR code display.
  • CMD Compatibility: Shell backend switched to cmd.exe for better compatibility with Node.js tooling on Windows.
  • Background Execution: The agent can launch long-running servers in the background and continue working concurrently.
  • Interactive Input: New send_process_input tool allows the agent to respond to interactive CLI prompts.

File Navigation

  • Built-in File Explorer: Browse and select project files directly from the left sidebar.
  • File Click to Analyze: Clicking any file automatically sends it to the agent for inspection.

Configuration

  • Universal Setup Wizard (CLI): tuki config --setup now covers all providers — Ollama, OpenRouter, Gemini, and Anthropic — in a single guided flow.
  • In-Chat Setup Wizard: The /setup command opens a step-by-step configuration modal inside the TUI without leaving the session.
  • Command Hints: Typing / in the input bar shows real-time command suggestions in the status bar.
  • Config Display: tuki config now shows a formatted summary table instead of raw JSON output.

Keyboard Shortcuts

  • Ctrl+S — Emergency stop: immediately halts the current agent execution.
  • Ctrl+B — Toggle the Live Console panel.
  • Ctrl+L — Clear the chat log.

Safety & UX

  • Input Lock: The input bar is visually locked while the agent is running, eliminating redundant error messages.
  • Bimodal Output: Shell results route to the Live Console; the agent's reasoning appears cleanly in the chat without ANSI noise.
  • Thinking Panel: Real-time reasoning display during inference, hidden automatically when complete.
v1.1.0

Infrastructure & Performance Update

Focused on multi-model orchestration and terminal execution stability.

Core & Architecture

  • Multi-Provider Support: Native integration for Google Gemini, Anthropic Claude, and OpenRouter.
  • Background Execution: Non-blocking shell tasks via subprocess management.
  • Context Optimization: Automatic output truncation for large terminal logs to prevent context saturation.
  • Project Exploration: Intelligent file filtering for get_project_tree, ignoring common build directories.

UI & Experience

  • Model Selector: Interactive modal for real-time model switching without restarting the session.
  • API Key Management: Secure input screens for provider credentials with persistent storage.
  • Process Control: Tools to list and terminate background processes by PID.
v1.0.0

Initial Release

Stable release of TukiCode's core autonomous agent architecture.

  • ReAct Reasoning: Implementation of the Think-Plan-Act inference cycle.
  • Terminal UI: Rich-based interface with real-time streaming output and mascot animations.
  • File System Tools: Secure CRUD operations for local file management.
  • Local AI: Full integration with Ollama for local-first, privacy-focused execution.