Zelos

Zelos is an autonomous AI coding agent extension for Visual Studio Code using the KIE API.

Zelos operates in an autonomous loop inside your workspace, utilizing advanced capabilities to analyze files, run tests, critique code, and automatically refactor or implement features based on your instructions.

💰 Extremely Cost-Effective Agentic Loops

Running autonomous AI agents typically consumes a massive amount of tokens due to continuous reasoning, context injection, and iterative tool steps. Zelos is built exclusively for the KIE API, which provides high-performance models at prices 30% to 80% cheaper than official APIs. This makes running complex autonomous agent loops highly affordable and accessible for everyday development.

---

✨ Features

• 🤖 Autonomous Agent Loop: Zelos executes multiple actions sequentially, reasoning through terminal outputs and file changes step-by-step until the task is complete.
• 🛠️ Rich Workspace Integration: The agent can perform real-world actions directly in VS Code through approval-guarded XML-based tools:
  • <create_file>: Create or modify files in the workspace.
  • <read_file>: Read file content for analysis.
  • <list_files>: List project structures.
  • <run_command>: Execute terminal commands (compiling, testing, formatting).
  • <visual_review>: Ask a separate visual model to review page UI, click buttons, type inputs, scroll, and verify functionality on a running Chrome instance.
• 📸 Multimodal Image Upload & Context:
  • Chat Image Attachment: Directly attach images in the Zelos sidebar panel. Images are uploaded to KIE's temporary storage, and their URLs are passed into the model's multimodal context.
  • Visual Review screenshots: The navigation review subagent automatically uploads screenshots of the debug Chrome browser using the KIE base64 upload API, facilitating visually guided page analysis.
  • Supports Gemini 3.5 Flash with custom endpoints and inline_data payloads.
• 🛡️ Secure Approval Workflows: Choose between:
  • Prompt mode (default): Review and approve every terminal command or file modification before it runs.
  • Auto-approve/Reject: Full execution automation or high-security lockouts.
• 🔍 Comprehensive Workspace Audit:
  • Architecture Reviews: Checks directories and structures against clean code paradigms.
  • Automated Code Review: Flags bugs, anti-patterns, and vulnerabilities.
  • Cognitive Complexity Audit: Evaluates functions and methods against a customizable cognitive complexity threshold (e.g., nesting, logical branches) and plans refactorings.
  • Self-Critique & Auto-Correction: Autonomously corrects detected flaws or failing tests by modifying the workspace and re-running the test suite.
• 🗜️ Token-Saving Conversation Compaction:
  • Zelos intelligently compacts verbose tool outputs and older message histories. This drastically reduces the context size for longer conversations, lowering API costs while maintaining reasoning memory.
• 🎨 Premium Chat Webview: A highly responsive, custom-styled Outfit/Fira Code interface with real-time streaming status updates, interactive tool execution logs, and easy configurations.
• 💰 Real-Time Credit Balance: Track your KIE API account credit balance directly inside the webview panel top bar with a click-to-refresh badge.

---

🚀 Installation & Requirements

Prerequisites

• VS Code ^1.80.0
• Node.js & npm (for compiling and executing actions)
• A valid KIE API Key from KIE.ai

Setup

1. Clone or download the repository into your VS Code extensions or development directory:

   git clone https://github.com/TrackZone1/Zelos.git

2. Navigate to the folder and install dependencies:

   npm install

3. Compile the extension:

   npm run compile

4. Press F5 in VS Code to launch a new Extension Development Host window with Zelos active.

---

⚙️ Configuration

Open VS Code settings (Ctrl+, or Cmd+,) and search for Zelos to configure:

Setting	Type	Default	Description
zelos.api.key	string	""	Your secret API key from KIE.
zelos.api.url	string	https://api.kie.ai	Base URL for the KIE API endpoints.
zelos.api.model	string	gpt-5-5	The LLM model used (e.g. gpt-5-5, gpt-5-codex).
zelos.commandApprovalMode	enum	prompt	Approval strategy for running terminal commands (prompt | acceptAll | rejectAll).
zelos.fileApprovalMode	enum	prompt	Approval strategy for modifying workspace files (prompt | acceptAll | rejectAll).
zelos.communicationLanguage	enum	English	Language used for conversational communication with the agent (English | French | Spanish | German | Italian | Portuguese | Japanese | Chinese).
zelos.codeLanguage	enum	English	Language applied for comments, variable naming, and documentation in the generated code (English | French | Spanish | German | Italian | Portuguese | Japanese | Chinese).
zelos.api.visualModel	string	gemini-3.5-flash	The model used for visual review and page navigation (e.g. gemini-3.5-flash, gpt-5-4).
zelos.chrome.selectedProfile	string	Default	Chrome profile directory to launch for debugging.

Alternatively, you can manage these settings directly inside the Zelos Webview Settings Panel (⚙️).

---

📖 Usage

💬 Regular Assistant Chat

1. Open the Zelos view in the VS Code Activity Bar (represented by the bot $(hubot) icon).
2. Input your API credentials in the settings panel (⚙️).
3. Send a message to start pair programming.
4. If approval modes are active, you will see interactive buttons inside the chat view asking to approve or reject actions (like running npm test or saving changes to a file).

🔍 Running Workspace Audits

1. Click the Audit button in the top bar.
2. Select your audit options:
   • Check directory layout & architecture
   • Perform code review & quality check
   • Run tests automatically (configurable test command)
   • Perform cognitive complexity analysis (with threshold)
   • Auto-critique & Self-correct issues
3. Click Start Audit and watch Zelos examine, test, critique, and self-heal your workspace autonomously.

🌐 Visual Review & Browser Control

1. Click the Browser button in the top bar to open the Chrome CDP control panel.
2. Select your desired Chrome Profile.
3. Click Launch Browser to launch a local Chrome instance on debug port 9222.
4. Run instructions requiring visual inspection or web navigation using the <visual_review url="..." instruction="..."></visual_review> tool block.
5. The Visual Review subagent will launch, navigate the page, perform clicks/scrolls, capture screenshots, upload them to KIE, and review the UI for bugs or layouts.

---

🧩 How it Works

Zelos bridges the gap between LLMs and local development environments by running an autonomous agent loop:

graph TD
    User([User Prompt / Audit]) --> AgentLoop[Agent Loop Start]
    AgentLoop --> SendAPI[Send history & context to KIE API]
    SendAPI --> GetResponse[Receive structured response]
    GetResponse --> CheckTools{Contains XML Tools?}
    
    CheckTools -- Yes --> Approve{Requires Approval?}
    Approve -- Approved --> Exec[Execute Tool local command/file]
    Approve -- Rejected --> FeedbackError[Feed rejection back to LLM]
    Exec --> Feedback[Feed result back to LLM]
    Feedback --> AgentLoop
    FeedbackError --> AgentLoop
    
    CheckTools -- No --> Render[Render final summary text]
    Render --> Done([Done])

Loading

1. Context Extraction: When starting a conversation, Zelos automatically injects the active editor's file content to provide immediate context.
2. KIE Codex API: Requests are sent to KIE's Codex endpoints (https://api.kie.ai/codex/v1/responses for gpt-5-5).
3. XML Tool Parser: The agent outputs structured XML tags. The extension captures and parses these commands, runs them locally, and returns the output directly to the LLM context.
4. Code-Leak Protection: If the agent leaks raw code in its text output (instead of writing it to files via tools), Zelos intercepts, prompts a system correction, and retries automatically to guarantee workspace consistency.

---

📄 License

This project is licensed under the GNU License. See the LICENSE file for details.