# What to use for ml engineering

> Training, serving and monitoring models — MLOps, feature stores, drift, ONNX/TensorRT. `model-evaluation` only covers scoring output; a computer-vision or MLOps capability had no home.

Source: https://tashan.sh/task/ml-engineering.html
Ranked by fit for the task, then how well it documents itself, then the tashan score
  (upkeep and freshness, gated by real adoption). Public evidence only — nothing paid can
  change a rank. Method: https://tashan.sh/methodology.html

## Ranked

| # | Capability | tashan score | Adoption evidence | Activity |
|---|---|---|---|---|
| 1 | [AI Context](https://tashan.sh/capability/pkg-vibgrate-cli.html) | 69 | 2k/wk | active |
| 2 | [Agent Nuvira](https://tashan.sh/capability/pkg-agent-nuvira.html) | 71 | 5k/wk | active |
| 3 | [Tuning Engines - Governed AI Runtime](https://tashan.sh/capability/pkg-tuningengines-cli.html) | 56 | 241/wk | active |
| 4 | [Gate402](https://tashan.sh/capability/pkg-gate402-mcp.html) | 53 | 234/wk | active |
| 5 | [Gyoshu](https://tashan.sh/capability/plugin-yeachan-heo-my-jogyo-gyoshu.html) | 49 | 240 ★ | stable |
| 6 | [Edgegate](https://tashan.sh/capability/pkg-edgegate-mcp.html) | 42 | 11/wk | active |
| 7 | [Compare CLI](https://tashan.sh/capability/pkg-compare-cli-mcp.html) | 40 | 66/wk | active |
| 8 | [Sparkrun](https://tashan.sh/capability/plugin-spark-arena-sparkrun-sparkrun.html) | 69 | 449 ★ | active |
| 9 | [Neurarch](https://tashan.sh/capability/pkg-neurarch-mcp.html) | 67 | 2k/wk | active |
| 10 | [Gerbil](https://tashan.sh/capability/pkg-tryhamster-gerbil.html) | 64 | 436/wk | active |
| 11 | [Gemini · houtini](https://tashan.sh/capability/pkg-houtini-gemini-mcp.html) | 63 | 819/wk | active |
| 12 | [Inference](https://tashan.sh/capability/pkg-bsvkey-inference-mcp.html) | 60 | 482/wk | active |
| 13 | [RASON](https://tashan.sh/capability/pkg-frontlinesystems-rason-mcp-server.html) | 39 | 78/wk | active |
| 14 | [Drift](https://tashan.sh/capability/pkg-use-tusk-drift-mcp.html) | 39 | 31/wk | active |

## What these numbers are not

- The tashan score measures upkeep, freshness and adoption. It is **not** a security
  verdict and **not** a measure of whether the capability works well.
- `not scored` means too little public evidence to rank, never that something is bad.
- The security audit is separate and free per capability, on each page above.