Project Roadmap

Project Roadmap

Gemma CLI is evolving from a sophisticated terminal wrapper into a fully autonomous engineering workstation. Below are the key milestones and features planned for future releases.

---

🤖 The Next step in Autonomy (Agentic Core)

• Goal-Stack Orchestration: Move beyond single-step tool chaining to high-level objective planning. You give a goal ("Refactor all scripts to use strict mode"), and Gemma autonomously plans, executes, and verifies subtasks.
• Gemma-First Backend Expansion: Transition the core workstation to support Gemma models from any provider (Local Ollama/vLLM, Cloud providers like Groq/Fireworks, or Google AI Studio).
• Hybrid Dual-Agent Backends: Allow the secondary reasoning layer in dual-agent pipelines to connect to any API (OpenAI, Anthropic, etc.), while keeping Gemma as the primary context-aware orchestrator.
• Multi-Agent Parallelism: Fan out multiple sub-agents using PowerShell background jobs to perform independent research or coding tasks simultaneously, then merge results.
• Self-Improvement Loop: Gemma analyzes her own tool-call failures and success patterns, proposing edits to her own ToolUseGuidance strings to become more accurate over time.

---

🧠 Memory & Context (The Second Brain)

• Persistent Workspace Contexts: Support for named workspaces (/workspace ProjectAlpha). Each has its own memory, pinned files, custom tools, and preferred model.
• Cross-Session Semantic Search: Archive every conversation and expose them via embedding search. Use /remember "that OAuth approach" to pull relevant code from weeks ago.
• History Pinning (/pin): Mark specific turns or code blocks as "never trim," allowing vital context to survive the Smart Trim algorithm indefinitely.
• ID-Catalogued History: Transition memory and history into a structured DB/JSON schema for more granular retrieval.

---

🛠️ Developer Experience (DX)

• Structured Diff-Based Editing: A patch_file tool that generates semantic diffs instead of overwriting files. Review colored hunks in the terminal and approve/reject individual changes.
• Tool Output Caching: Cache results for idempotent tools (readfile, searchdir) within a session to save API costs and increase speed.
• Tool Marketplace & Auto-Installer: A community manifest hosted on GitHub. Install new capabilities instantly via /tools install [name].
• Usage Analytics & Insights: A local JSON log tracking tool success rates and frequency. Surface data via the /insights command.

---

🎨 Multimodal & Specialized Tools

• Arduino CLI Integration: Direct hardware control and firmware deployment through the CLI.
• Voice & Audio (STT/TTS): Real-time speech-to-text input and text-to-speech output for eyes-free operation.
• Advanced Creativity: Expanding the Imagen toolset to support video generation and high-fidelity restoration.
• Persona Engine: Quick-load pre-configured expertise models (e.g., /persona GameDev, /persona SecurityExpert) via persona.json.

---

🐚 Power User Utilities

• /undo: Pop the last turn pair (user + model) from history to quickly correct a prompt or model error.
• /export: Save the current conversation as a clean, formatted Markdown or HTML file for documentation.
• Smart Home Tools: Specialized modules for interacting with IoT environments and local automation hubs.

---

Note: This roadmap is community-driven. If you have an idea for a "Slam Dunk" tool, please contribute via a Pull Request.