WUMBOLABS / PROJECT RECORD
Monolith
Local-first AI workbench for testing, comparing, and evaluating local LLMs on real hardware.
Try Monolith
Monolith is the local AI workbench layer on WumboLabs: a local web UI for model tracking and future LLMGauge import workflows. LLMGauge remains the flagship public evidence tool.
It is a technical public alpha. Setup is still manual, and it is intended for users comfortable with Linux, Python, local model files, GGUF, and llama.cpp-style workflows.
1. Open the repo
Start from the GitHub repository and read the current public alpha setup notes.
2. Clone it locally
Monolith runs as a local web UI. It is not a hosted service or production deployment target.
3. Run setup checks
Use the setup diagnostics to confirm the local environment before adding real model profiles.
4. Start the local UI
The current local interface runs at http://127.0.0.1:8765/.
Current Development Note
Recent Monolith development is focused on importing LLMGauge artifacts into the local workbench: import tables, artifact metadata parsing, importer architecture, listing pages, detail pages, and sidebar UI polish.
That work is active development beyond the current public alpha release metadata, so it should be read as current project direction rather than a packaged release claim.
Technical Public Alpha · alpha v0.11.12
At a Glance
Capabilities
- Local GGUF model profile tracking
- Generated chat profiles from configured local models
- Basic chat and test run logging
- Prompt and response metadata capture
- Token count and speed tracking
- VRAM and performance fields
- Local eval prompt suite browsing and imports
- Context-scaling evaluation scaffolding
- Agent-backend evaluation scaffolding
- Agent Lab proposal, review, and read-only workflow scaffolding
- Setup diagnostics through the UI and terminal-side checks
- Workstation monitoring with graceful fallback for optional metrics
Current caveats
- Technical public alpha
- Manual setup
- No one-command installer yet
- No bundled llama.cpp install
- No GPU driver, CUDA, ROCm, or system runtime setup
- No production multi-user or authenticated deployment model yet
- No open-source license selected yet
Technical details
Canonical Local Usage
Current local web UI:
http://127.0.0.1:8765/Canonical start command:
python scripts/run_webui.pyAudience
Technical users comfortable with Linux, Python virtual environments, llama.cpp, GGUF models, YAML configuration, and manual setup.
License Status
Source-available; no open-source license selected yet
Current Roadmap
The roadmap below is synced from Monolith's canonical docs/ROADMAP.md public website block.
Monolith public roadmap
Monolith is a technical public alpha for local LLM workbench development, model testing, and practical evaluation on real hardware.
Current phase:
alpha v0.11.12— repo-local bootstrap script and guided setup preparation- clean-clone install validation
- setup diagnostics
- dependency and runtime detection
- safer model onboarding
Next milestone:
alpha v0.11.13— guided setup wizard
Upcoming phases:
- guided first-run setup flow
- clearer llama.cpp/runtime detection
- safer model profile editing
- improved model onboarding
- stronger setup doctor output
- cleaner local evaluation workflow
- better comparison views for model runs and context tests
Future direction:
- eval scoring workflow
- comparison dashboard
- global search or command palette
- run notes and annotations
- agent-backend readiness dashboard
- workstation monitoring improvements
- future CLI/TUI companion
Current caveats:
- Monolith is a technical public alpha.
- Monolith is source-available; no open-source license has been selected yet.
- Setup is still manual, though repo-local bootstrap support is being added.
- llama.cpp is not bundled or installed automatically.
- GPU drivers, CUDA, ROCm, and system runtimes are not installed or modified by Monolith.
- There is no one-command installer yet.
- There is no production multi-user/auth deployment model yet.