Skip to content
NIKI

Quick Start

Your First Monitored Job

Once NIKI is installed and configured, monitoring a job is simple:

Terminal window
niki watch -- python train.py --epochs 1000

That’s it! NIKI will:

  1. Start your command in the background
  2. Capture stdout/stderr
  3. Extract key metrics when it finishes
  4. Send you a notification with an AI-powered summary

Basic Usage

Watch a Command

Terminal window
# Basic syntax
niki watch -- <your command>


# With a custom name
niki watch --name "Training ResNet" -- python train.py


# With timeout
niki watch --timeout 3600 -- python long_job.py

Attach to Running Process

Already have a job running? Attach to it:

Terminal window
# By PID
niki attach 12345 --name "Alignment job"


# Find running Python processes
ps aux | grep python

Check Status

Terminal window
# See running/recent tasks
niki status


# Query history
niki history --last 7d
niki history --name "train*"

What You’ll Receive

When your job completes, you get a notification like this:

✅ Training Complete (2h 34m)


📍 workstation
📊 Loss: 0.0023 (↓ from 0.847)
🎯 Accuracy: 98.7%
💾 Best checkpoint: epoch 847


Reply "logs", "rerun", or ask a question

Jupyter Integration

For Jupyter notebooks:

%load_ext niki


%%niki_watch preprocessing
import scanpy as sc
adata = sc.read_h5ad("data.h5ad")
sc.pp.filter_cells(adata, min_genes=200)
sc.tl.umap(adata)

The cell runs normally, and when it completes, you get a notification with any plots captured.

Reply Commands

You can interact with NIKI via Telegram/Slack:

  • “logs” - Get the last 50 lines of output
  • “rerun” - Rerun the last job
  • “show loss plot” - Ask for specific information
  • “what was the final accuracy?” - Natural language queries

Next Steps