v0.2.1a

FOR IMMEDIATE RELEASE:

• Text is now selectable in message bubbles

• Smooth scroll (per-pixel instead of per-item)

• Scroll no longer forces you to the bottom during generation

• UPDATE is now silent (no visible user message in chat)

• Chat title locks after first message (no more overwriting on follow-ups)

• Edit no longer wipes the reasoning trace panel

• SESSION RESET message removed from chat (trace only)

• MONOLITH label hides on hub, shows everywhere else

v0.2a

Added (+)
(+) Overseer real-time kernel/engine trace dashboard with SQLite persistence
(+) Operator system for saving/restoring full workspaces (modules, configs, history, models)
(+) UIBridge for UI-to-UI signaling separate from kernel commands
(+) VizTracer profiling hooks inside kernel execution
(+) Thinking-mode + ephemeral runs in LLM engine
(+) Engine-scoped token/trace/usage signals for multi-engine awareness
(+) Kernel commands for restoring history and updating context limits
(+) Hub page as default workspace launcher
(+) Persistent Overseer trace database
(+) Addon registry extensions + new builtin factories

---

Changed (~)
(~ ) STOP / generation lifecycle hardened to prevent ghost output
(~ ) Chat system rewritten with per-message edit/delete/regenerate
(~ ) LLM system-prompt pipeline locked to MASTER_PROMPT + behavior tags
(~ ) Config loader now strips deprecated override fields
(~ ) Model loader now reads true training context window and clamps ctx limits
(~ ) Dock now emits structured trace logs for Overseer
(~ ) Bootstrap wiring updated for Overseer + UI bridge + clean shutdown
(~ ) Audio/Vision engines moved closer to LLM parity

---

Removed (-)
(-) Direct user override of full system prompt via config
(-) Legacy monolithic chat widget implementation
(-) Silent generation-state leakage during edit/regenerate
(-) Hardcoded UI navigation paths bypassing UIBridge

v0.1.0-alpha

Monolith v0.1.0-alpha

First public release.

What is Monolith?

A local-first AI workstation that runs LLMs, Stable Diffusion, and audio generation through a modular kernel architecture.

Stop chatting with AI. Start commanding it.

---

What Works

✅ LLM chat (GGUF models via llama.cpp)
✅ Stable Diffusion image generation
✅ Audio generation (AudioCraft)
✅ Persistent chat sessions
✅ Modular addon system
✅ Dark mode interface

Known Issues

⚠️ This is alpha software
⚠️ Limited error handling
⚠️ Documentation incomplete
⚠️ Windows only (Linux/Mac untested)
⚠️ Rough edges everywhere

Installation

Windows:

1. Download and extract
2. Run install.bat
3. Run start.bat

See README for details.

Requirements

• Python 3.10+
• CUDA GPU recommended
• ~10GB disk space for models