Register a local model. Run it through one harness. Offline-first CLI + TUI for people who keep models on disk (sklearn, embeddings, Ollama/GGUF/HF, vision, more) and want one interface instead of five tools.
PyPI · macOS / Linux (v1) · Apache 2.0
I kept wiring the same “detect model → run the right commands → maybe serve HTTP” glue for different local stacks. Ollama, Gradio, and BentoML each solve a slice. I wanted a thin registry + harness layer that stays offline-first and plugin-shaped. So I shipped everyharness as a CLI/TUI on PyPI.
| Piecing Ollama / Gradio / notebooks | everyharness | |
|---|---|---|
| Interface | Different per model type | One CLI / TUI |
| Offline | Easy to accidentally hit the network | EVERYHARNESS_OFFLINE=1 hard-blocks outbound |
| Plugins | Ad hoc scripts | everyharness-* packages + plugin init |
| Honesty | Often “works for everything” marketing | Alpha table — what’s solid vs thin |
Core
- Model registry (local files, Hugging Face, Ollama, Python callables)
- Harness selection by kind / URI (override with
--type) - Same flow:
add→run→serve→train(train depth varies) - Textual TUI: bare
everyharness - Agent prompt pack:
everyharness uiwrites files for a coding agent (doesn’t build the UI itself)
Maturity (alpha — be honest)
| Area | Status |
|---|---|
| Tabular (sklearn/joblib) | Solid — predict, evaluate, explain, HTTP serve |
| Embeddings | Usable |
| LLM | Thin wrapper (Ollama; optional GGUF / HF) |
| Vision | Classify only |
| Diffusion | CLI generate only |
| Speech | Transcribe only (install whisper yourself) |
| Computer | Experimental |
| Community plugins | None curated yet |
Pickles need --trust-pickle. Prefer joblib from sources you trust.
pip install everyharness
# optional: pip install 'everyharness[tabular]' / '[llm]' / '[all]'
everyharness add ./model.pkl --trust-pickle
everyharness add embeddings:demo --type embeddings
everyharness list
everyharness run --trust-pickle <id> predict --input '[[1.5, 0.5]]'
everyharnessDev: uv sync --all-extras --dev && uv run pytest -q
Python · Textual TUI · plugin host · optional extras for tabular / LLM / vision · Apache 2.0