Roadmap

This document outlines the development roadmap for Morpheum.

Morpheum v0.1: The “Matrix Milestone” (Completed)

This initial phase focused on bootstrapping the project and establishing the basic infrastructure for a collaborative, Matrix-based workflow.

1. Matrix Room Setup (Done)

  • A dedicated Matrix room for the Morpheum project was created and serves as the central hub for all project-related communication.

2. Bot Development: The “Morpheum Bot” (Done)

The core of this milestone was the development of the “Morpheum Bot” to act as the primary interface between developers and the GitHub repository.

  • Initial Proof of Concept (Done):
    • The Gemini CLI was successfully forked and integrated as a library to bootstrap the bot’s development. This allowed for an initial proof of concept and has since been replaced by a direct integration with local LLMs, marking this as a successful bootstrapping step.
  • Basic Bot Infrastructure (Done):
    • A basic bot framework using TypeScript and the matrix-bot-sdk has been set up.
    • The bot is configured to join the project’s Matrix room and respond to basic commands.
  • GitHub Integration (Done):
    • A comprehensive GitHub integration has been implemented via CopilotClient that can create issues, manage pull requests, track GitHub Copilot sessions, and perform repository operations using both REST and GraphQL APIs.
  • Agent Integration (Done):
    • The bot’s agent is fully operational and invoked by mentioning the bot’s name (@botname: <prompt>). The underlying logic handles a wide range of commands including SWE-Agent operations, gauntlet testing, task management, and GitHub operations.

3. Workflow Transition (Done)

  • Dogfooding (Done): The project is actively using the bot for its own development (“dogfooding”) with Matrix-based workflows, GitHub Copilot integration, and automated task management.
  • Documentation (Done): Project documentation has been restructured to use directory-based systems (eliminating merge conflicts) and is continuously updated to reflect the current workflow state.

Morpheum v0.2: Agent Advancement (Current Focus)

The primary goal for the current phase is to enhance the intelligence, reliability, and capabilities of the core SWE-Agent. This involves rigorous testing, targeted improvements, and integration with a wider range of tools.

1. Agent Evaluation and Improvement

  • Gauntlet Testing:
    • Systematically run the evaluation gauntlet (GAUNTLET.md) against a variety of local and proprietary models to establish performance benchmarks.
    • Analyze the results to identify common failure points and areas for improvement in the agent’s planning and execution logic.
  • Prompt Engineering:
    • Iteratively refine the system prompts in prompts.ts based on gauntlet results to improve the agent’s reasoning and tool-use capabilities.
  • Self-Correction and Learning:
    • Investigate and implement mechanisms for the agent to learn from its mistakes. This could involve feeding back summaries of failed tasks into its context or developing a more sophisticated self-correction loop.

2. Enhanced Tooling and Environment

  • OpenAI API Integration (Done):
    • Complete integration with the OpenAI API has been implemented, allowing the bot to leverage models like GPT-4 for tasks that require more advanced reasoning. The bot supports both OpenAI and Ollama backends with automatic fallback.
  • Jail Environment Enhancements (Done):
    • A comprehensive jailed development environment has been implemented supporting Nix-based containerization, Docker integration, and TCP-based agent communication for secure, isolated code execution.

3. Workflow and Usability

  • Matrix Interface:
    • Improve the user experience in the Matrix room by providing more structured output, better error reporting, and more intuitive commands.
  • Multi-Agent Collaboration:
    • Begin experimenting with multi-agent workflows, where different agents with specialized skills can collaborate on a single task.