ELVES

Product Introduction

1. Definition: ELVES is an open-source AI Agent Orchestration platform developed as a Tauri-based desktop application. It acts as a specialized management layer for local AI coding agents, specifically designed to coordinate high-performance CLI-based agents like Claude Code and Codex within a unified, isolated development environment.

2. Core Value Proposition: ELVES addresses the fragmentation and operational friction inherent in using multiple AI coding agents. By providing a worktree-isolated workspace, persistent memory across sessions, and a centralized "Ship It" deployment flow, it transforms chaotic agent interactions into a structured, production-ready development pipeline. It is optimized for developers who require AI agent orchestration that respects local git states and maintains high-context continuity.

Main Features

1. Worktree-Isolated Workspaces: ELVES utilizes native Git worktree functionality to create fully isolated working directories for every task. When a user inputs a natural language prompt, the application automatically spawns a new worktree and branch. This ensures that parallel AI tasks never interfere with the main codebase or each other, allowing for safe experimentation and side-by-side comparison of agent outputs.

2. WebGL-Accelerated Embedded Terminal: The platform features a high-performance PTY (Pseudo-Terminal) terminal rendered via WebGL for GPU-accelerated performance. This terminal is specifically wired to interface with Claude Code or Codex. It includes a built-in toolbar, advanced search overlays, Unicode support, and automatic session persistence, enabling developers to watch agents execute commands and write code in real-time without losing progress upon restarts.

3. Persistent SQLite-Backed Memory: ELVES implements a long-term memory system using a local SQLite database enhanced with FTS5 (Full-Text Search). This system stores agent decisions, project context, and user preferences. It employs a relevance decay algorithm where critical architectural decisions remain accessible while transient "noise" fades over time, ensuring agents remain context-aware across different sessions and projects.

4. Advanced Task Analysis & Mode Selection: Before spawning an agent, ELVES performs a classification of the task's complexity. It determines the optimal runtime configuration and selects between "solo" or "team" execution modes. This pre-orchestration phase ensures the underlying engine (Claude Code or Codex) is properly prepared for the specific technical requirements of the task.

5. Integrated Ship It Flow: The application streamlines the transition from AI-generated code to production. It features a visual merge strategy picker that supports standard git operations including merge, rebase, and squash. Upon shipping, the tool automatically extracts relevant memories for future context, cleans up the isolated worktree, and deletes the temporary branch to maintain repository hygiene.

Problems Solved

1. Pain Point: Context Switching and Workspace Pollution. Developers often struggle with AI agents making broad changes to a single working directory, leading to "dirty" git states. ELVES solves this by physically isolating every agent action into separate worktrees.

2. Pain Point: Volatile Agent Memory. Standard CLI agents often lose context between different terminal sessions. ELVES provides a persistent data layer that bridges the gap between individual agent runs, making the AI "smarer" the more it is used within a specific codebase.

3. Target Audience: Software Engineers, DevOps Architects, Full-Stack Developers using AI-driven workflows, and Open-Source Contributors. It is particularly valuable for teams integrating Claude Code or Codex into their daily CI/CD and local development routines.

4. Use Cases:

• Implementing complex features (e.g., "Add OAuth authentication") where multiple files must be edited and tested simultaneously.
• Running parallel refactoring experiments without affecting the primary development branch.
• Analyzing telemetry and token usage across different AI models to optimize development costs and performance.
• Managing MCP (Model Context Protocol) servers and project-specific context through a visual GUI rather than manual configuration files.

Unique Advantages

1. Differentiation: Unlike web-based AI IDEs or simple wrapper scripts, ELVES is a native desktop application that leverages local system primitives like git worktrees and PTY terminals. This provides a "neo-brutalist" interface that prioritizes speed, local security, and deep integration with the developer's existing local toolchain.

2. Key Innovation: The combination of "Runtime Agnosticism" and "Physical Isolation." ELVES allows users to switch between different AI engines (Claude Code vs. Codex) while maintaining the exact same terminal interface, memory context, and git-safe environment. This decoupling of the orchestrator from the AI engine provides future-proof flexibility as new coding models emerge.

Frequently Asked Questions (FAQ)

1. How does ELVES ensure git safety when using AI agents? ELVES uses Git Worktrees to create entirely separate physical directories for each AI task. This means the AI agent never touches your primary working directory until you explicitly choose to "Ship It." If an agent produces undesirable results, you can discard the worktree without any impact on your main branch or uncommitted changes.

2. Can ELVES work with any AI coding agent? Currently, ELVES is optimized for Claude Code and Codex. Its runtime-agnostic architecture is designed to support additional CLI-based AI agents in the future, providing a consistent interface and shared SQLite memory regardless of the underlying LLM engine.

3. Is ELVES available for Windows and Linux? ELVES is currently optimized for macOS. However, the roadmap includes support for Linux and Windows systems. Since it is built on the Tauri framework, cross-platform compatibility is a core objective for upcoming releases.

4. Does ELVES track my AI token usage and costs? Yes, ELVES includes a Usage Insights Dashboard. This feature provides real-time telemetry, including token usage categorized by model, daily activity charts, and project-specific breakdowns. It even generates AI-powered narrative reports to help you understand your productivity and agent efficiency.