PRysm

An open source AI-powered GitHub PR review bot with memory.
It watches for pull requests, reads the diffand posts a detailed code review
directly on GitHub and gets smarter with every PR it reviews.

how it works

1. someone opens a PR on your repo
2. GitHub sends a webhook to PRysm
3. PRysm fetches the diff and checks its memory for past review patterns on this repo
4. the diff + memory context is sent to Llama via Groq
5. the AI reviews it and PRysm posts inline comments directly on the diff lines
6. the review is saved to memory so future PRs on this repo benefit from past context

---

features

• inline comments : feedback posted directly on the changed lines in the diff
• per-repo memory : PRysm remembers past reviews for each repo separately and gets smarter over time
• chunked review : handles large PRs by splitting the diff into chunks
• smart file filtering : skips lock files, binaries, minified filesand build artifacts
• auto retry : if the LLM returns invalid JSON, PRysm automatically retries with a correction prompt

---

memory

PRysm uses Hindsight to remember past PR reviews. Each repo gets its own isolated memory bank. Over time PRysm learns the patterns, recurring issuesand code style of your codebase and uses that context when reviewing new PRs.

the graph above shows PRysm's memory bank for a test repository after 12 PRs each node is a memory, connected by semantic, temporaland entity relationships.

---

prerequisites

before you start, you need:

• a GitHub personal access token needs repo scope. get it at github.com/settings/tokens
• a Groq API key free at console.groq.com
• a webhook secret just make up any random string, you'll use it in both your .env and GitHub webhook settings
• a Hindsight API key free at ui.hindsight.vectorize.io (for memory)

---

setup

method 1 use the hosted version (easiest)

no setup needed. PRysm is already deployed and running, just point your GitHub webhook at it.

step 1 go to your GitHub repo -> settings -> webhooks -> add webhook

step 2 fill in the form

field	value
payload url	https://prysm-pr-review-agent.onrender.com/webhook
content type	application/json
secret	make up any random string and save it
events	select "let me select individual events" -> check pull requests only

---

method 2 clone the repo

step 1 clone and enter the project

git clone https://github.com/akshattalwar001/PRysm.git
cd PRysm

step 2 create a virtual environment and install dependencies

python -m venv venv

# on mac/linux
source venv/bin/activate

# on windows
venv\Scripts\activate

pip install -r requirements.txt

step 3 create your .env file

open .env and fill in your keys:

GITHUB_TOKEN=your_github_personal_access_token
GROQ_API_KEY=your_groq_api_key
WEBHOOK_SECRET=your_webhook_secret
HINDSIGHT_API_URL=https://api.hindsight.vectorize.io
HINDSIGHT_API_KEY=your_hindsight_api_key

step 4 start the server

uvicorn main:app --reload --port 8000

server will be running at http://localhost:8000

---

method 3 docker

step 1 pull the image

docker pull akshattalwar/prysm-pr-review-agent:latest

step 2 create your .env file with your keys (same as method 2)

step 3 run the container

docker run -p 8000:8000 --env-file .env akshattalwar/prysm-pr-review-agent:latest

or with docker compose:

docker compose up --build

docker image also available at: https://hub.docker.com/repository/docker/akshattalwar/prysm-pr-review-agent

---

setting up the github webhook

once your server is running (and exposed via ngrok or a real server)

step 1 expose your local server (if running locally)

ngrok http 8000

copy the https:// URL ngrok gives you.

step 2 go to your GitHub repo -> settings -> webhooks -> add webhook

step 3 fill in the form

field	value
payload url	https://your-url.ngrok.io/webhook
content type	application/json
secret	same value as your WEBHOOK_SECRET in .env
events	select "let me select individual events" -> check pull requests only

step 4 click "add webhook" PRysm will now automatically review every PR opened on that repo.

---

project structure

PRysm/
├── main.py              
├── webhook_handler.py   
├── github_client.py     
├── context_builder.py   
├── llm_client.py        
├── comment_poster.py    
├── config.py            
├── requirements.txt     
├── .env           
├── Dockerfile           
└── docker-compose.yml   

---

License

MIT see LICENSE