PerformAI is a modular Python tool that connects to a Kubernetes cluster, analyzes CPU and memory usage of workloads using Prometheus metrics, and generates optimization recommendations using a local LLM (like Mistral 7B via Ollama) or hosted LLMs compatible with OpenAI API.
- Targets only specific namespaces
- Pulls resource usage from Prometheus over configured lookback duration
- Compares with actual CPU/Memory requests & limits
- Uses local or hosted LLM compatible with OpenAI API to get recommended adjustments
- Python 3.9+
- Prometheus endpoint accessible (you can follow Prometheus Setup to setup prometheus via Jupyter Notebook)
kubectlandocconfigured and authenticated- Ollama running locally with the
mistralmodel:ollama run mistral
- Optionally, hosted LLM URL, Model name, API Token
git clone https://github.com/yourusername/performAI.git
cd performAI
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txtSet the environment variables
export NAMESPACES="namespace1,namespace2,another-namespace"
export PROMETHEUS_URL="https://your-prometheus.example.com"
export LOOKBACK_DURATION="7d"
export CHUNK_SIZE="10"
export DEBUG="True"
For local LLM:
python main.pyFor hosted LLM:
USE_LOCAL_LLM=false \
LLM_HOSTED_URL=https://your-llm-host.com \
LLM_API_TOKEN=your-token-here \
LLM_MODEL=model-name \
python main.pyFor web UI
python app.pyserved on localhost:5000
If your cluster does not yet have Prometheus installed or configured, you can use the interactive Jupyter notebook provided to set it up:
python3 -m venv venv-setup
source venv-setup/bin/activate
pip install notebook
jupyter notebook setup/prometheus-setup.ipynbThe notebook will:
- Detect your current namespace and OpenShift domain
- Install Prometheus Operator on current namespace
- Create the necessary RBAC and scrape configurations
- Deploy Prometheus CR
- Provide an external route to access Prometheus UI
Note: You need
ocand sufficient cluster permissions to run this notebook on OpenShift.
| Variable | Description | Default |
|---|---|---|
NAMESPACES |
Comma-separated list of Kubernetes namespaces to monitor. | threescale |
PROMETHEUS_URL |
URL of the Prometheus instance to query. | `` |
LOOKBACK_DURATION |
The time window to query Prometheus for metrics. | 30d |
CHUNK_SIZE |
The size of workload chunks sent to the LLM for analysis. | 5 |
DEBUG |
Enable debug logging (set to "true" for verbose output). | false |
USE_LOCAL_LLM |
Whether to use local Ollama (true or false) |
true |
LLM_HOSTED_URL |
Base URL of the hosted LLM endpoint | (required if hosted) |
LLM_API_TOKEN |
API token to authenticate with hosted LLM | (required if hosted) |
LLM_MODEL |
Model name to use (e.g., mistral, gpt-3.5-turbo) |
mistral |
These environment variables control which LLM is used and how it's accessed. When using a hosted LLM, ensure the URL and token are valid for the OpenAI-compatible /v1/chat/completions endpoint.
performai/
├── config.py
├── collect.py
├── prompts.py
├── llm.py
├── utils.py
templates/
├── index.html
├── processing.html
├── results.html
app.py
main.py
setup/
└── prometheus_setup.ipynb
MIT