What's up? What's next.

NIKI

Your intelligent code monitor

Get intelligent notifications on Telegram or Slack when long-running jobs complete. With context that actually matters.

terminal
$ niki watch -- python train.py
Running in background...
# You'll be notified when done
โ†’
โšก NIKI now
โœ… Training Complete
2h 34m
Loss: 0.0023
Accuracy: 98.7%
Best epoch: 847
๐Ÿ’ฌ Reply to ask questions

Try it yourself

Type a command and see what happens when it finishes

terminal — niki demo
$ niki watch -- โ–‹

Press Enter to run command...

๐Ÿ“ฑ Telegram Notification
โœ“ Command (2h 34m)
๐Ÿ’ฌ Reply to ask questions or rerun

Ask questions. Get answers.

NIKI doesn't just notify โ€” it understands your code and helps you fix it

โšก NIKI Conversation

What caused the error?

The job failed due to CUDA out of memory at epoch 234. Your batch size (64) is too large for the 12GB GPU.

๐Ÿ’ก Suggested fix:

python train.py \
  --batch_size 32 \
  --epochs 1000

Run it

โœ… Started job #4521

I'll notify you when it completes.

Try asking: "What was my best epoch?" "Show the loss plot" "How much GPU memory used?"

More than just notifications

NIKI understands your code and helps you fix it

๐Ÿง 

Ask Anything

Query your jobs naturally. Get answers, not log files.

"What was my best epoch?"
"Show the loss plot"
"How much GPU memory used?"
๐Ÿ”ง

Understand Errors

AI explains why your job failed and suggests fixes.

โŒ CUDA out of memory
โ†“
"Batch size 64 is too large. Try --batch_size 32"
๐Ÿ”„

Remote Control

Edit, confirm, and rerun commands from your phone.

You said: "Rerun with lr=0.001"
python train.py --lr=0.001
โœ“ Run Edit
๐Ÿ“Š

Smart Summaries

Get the metrics that matter. Not 10,000 lines of logs.

10,000 lines
of logs
โ†’
Loss: 0.023
Acc: 98.7%

Designed with Safety in Mind

We built NIKI with a simple principle: do one thing well, and do it safely

โœ…

You Confirm

Every command requires your explicit approval before running. NIKI proposes โ€” you decide.

๐Ÿ 

Self-Hosted

Runs on your server. Your API keys. Your bot tokens. No data leaves your machine.

๐ŸŽฏ

Focused by Design

NIKI does one thing and does it well. No browser control, no broad system access, no surprises.

๐Ÿ“–

Fully Open Source

Audit every line of code. No black boxes. No hidden behavior.

github.com/genecell/niki

Your workflow, transformed

Stop babysitting your terminal

๐Ÿ˜ฉ Without NIKI
9:00 Start job
9:15 SSH, check... still running ๐Ÿ”„
9:30 SSH, check... still running ๐Ÿ”„
9:45 SSH, check... still running ๐Ÿ”„
10:00 SSH, check... still running ๐Ÿ”„
... (repeat for 2 hours)
11:30 SSH, check... DONE!
11:31 Scroll through 10,000 lines of logs...
11:35 Finally find the accuracy number
โฑ ~30 minutes wasted checking + digging through logs
๐Ÿ˜Ž With NIKI
9:00 Start job with niki watch
9:01 Close laptop, go get coffee โ˜•
... (live your life)
11:30 ๐Ÿ“ฑ Phone buzzes!
โœ“ Training complete (2h 30m)
Loss: 0.0023 • Accuracy: 98.7%
Best checkpoint: epoch 847
11:31 Reply: "show loss plot" ๐Ÿ“Š
โฑ 0 minutes wasted — got results instantly on your phone

Everything you need

Built for developers running long jobs on servers, clusters, and notebooks

๐Ÿง 

AI-Powered Summaries

Claude extracts key metrics from logs — loss, accuracy, error messages. No more scrolling through thousands of lines.

๐Ÿ“ฑ

Telegram & Slack

Get notified on your phone or team channel. Bidirectional — reply to ask questions or rerun jobs.

๐Ÿ”ฌ

Domain-Aware

Understands ML training, bioinformatics pipelines, ETL jobs. Shows the metrics that matter for each domain.

โšก

2-Minute Setup

pip install niki + interactive wizard. No complex configuration or infrastructure.

๐Ÿ““

Jupyter Integration

%%niki_watch magic monitors notebook cells. Captures plots automatically.

๐Ÿ”“

100% Open Source

MIT licensed. Run on your own infrastructure. Your data stays yours. No vendor lock-in.

๐Ÿ”’
Security First
Whitelist, rate limits, audit logs
๐Ÿ–ฅ๏ธ
Multi-Server
Monitor HPC, workstations, cloud
๐ŸŽฏ
Priority Levels
P0 (urgent) to P2 (low)
๐Ÿ—ฃ๏ธ
Voice Support
TTS announcements (optional)

Up and running in 2 minutes

No complex setup. Seriously.

1

Install NIKI

pip install niki[all]
2

Configure (interactive wizard)

niki config init
3

Start the daemon

niki daemon start
โœ“

Done! Monitor your first job

niki watch -- python your_script.py