This is a reading companion for the Developer Guide. Its only job is to help you understand what the guide is talking about. You will not learn to write code from scratch here — you will learn just enough to recognise what each tool, word, and step means, then go back and follow the guide.
The trick is to let an AI chat walk you through the practical setup. Every section below gives you a ready-to-paste prompt scoped to one specific task: installing Python, opening a terminal, cloning a repo, running a Streamlit app, and so on. Streamlit is the simplest path for most consultant tools; if you are building a non-Streamlit React/FastAPI app, start from the custom template linked in the Developer Guide.
Is Streamlit approved for ERM use?
Short answer: Streamlit is a supported application framework within the AI Marketplace deployment pattern. That is not the same as approving Streamlit as a standalone ERM application or SaaS service. The Developer Guide is implementation guidance, not ERM's approved-software register. The guide refers to the open-source Python framework used inside an AI Marketplace app; it does not approve Streamlit Community Cloud, another externally hosted Streamlit service, or arbitrary standalone deployments.
| What the Developer Guide covers | What it does not authorise |
|---|---|
| A Python app built from the supported template, packaged and deployed in ERM's AWS environment, and accessed through the AI Marketplace with ERM sign-in and platform controls. | Publishing the app to Streamlit Community Cloud, using an independent external Streamlit host, making a public deployment, or bypassing AI Marketplace authentication and deployment controls. |
Think of Streamlit here as a library inside the application, not as a separate SaaS product that ERM users sign into. The framework being available in the AI Marketplace delivery pattern is also not blanket approval for every application, dataset, integration, or use case. Each app must still follow the Marketplace's repository, SSO, hosting, security, data-handling, testing, and deployment requirements, with app-specific review where required.
If your app will process client confidential or sensitive data, connect to another system, or needs restricted access, do not assume the framework choice settles the approval question. Follow the AI security and client data guidance and raise the app-specific question with the relevant platform or governance owner.
What to tell the Service Desk
First confirm whether they mean Streamlit Community Cloud or the open-source framework inside an AI Marketplace-hosted app. You can send this wording:
Quick cheat sheet
One-line definitions of every word the Developer Guide throws at you, grouped by topic. Use this as a glossary while you read — these are the handful of things you need to recognise.
Python, Git, and Visual Studio Code should already be in your ERM self-service catalog and install automatically with no approval required. If one of those apps is missing from your catalog, raise a Technology Support request.
Python and your app
| Word | What it actually is |
|---|---|
| Python | A programming language. You write your tool's code in Python |
| Streamlit | The simplest app framework for most consultant tools. It turns a Python file into a live webpage — no HTML needed. Your main file is typically app.py. |
| React + FastAPI | The custom-app path for non-Streamlit cases that need a richer UI, protected backend routes, file processing, or workflow control. Use the ai-marketplace-custom-app-template. |
| Poetry | A helper that manages which Python libraries your app needs |
Commands you will type
| Command | What it does |
|---|---|
poetry install | Install all the libraries your app depends on (run from inside your repo folder, once after cloning) |
poetry add package-name | Add a new library to your app (replace package-name). Use this instead of editing files by hand |
poetry run streamlit run app.py | Start your app on your own computer so you can test it in a browser at http://localhost:8501 |
Ctrl + C | Stop your app when you are done testing (press this in the terminal where it is running) |
Git and GitHub
| Word | What it actually is |
|---|---|
| Git | A tool that tracks changes to your code and syncs them with GitHub |
| GitHub | A website where code is stored — like Google Drive, but for code projects. You need membership in ERM's GitHub org; use the GitHub access request instructions in the Developer Guide. |
| Repo (repository) | A single project folder on GitHub |
| Org (organisation) | A group on GitHub that owns repos. Marketplace app repos should usually be in ERMInc. |
| Template repo | A starter project you copy to begin your own. Streamlit currently uses capability-tool-template, but new app repos should still be named ai-marketplace-<name>. Custom React/FastAPI apps use ai-marketplace-custom-app-template. |
| Main branch | The official version of the code on GitHub |
| Clone | Download a copy of a GitHub repo to your own computer |
| Commit | Save a change you have made, with a short message describing it |
| Push | Upload your saved changes from your computer to GitHub |
| Pull | Download the latest changes from GitHub to your computer |
Commands you will type
| Command | What it does |
|---|---|
git config --global user.name "Your Name" | First-time setup — tell Git who you are so your commits are labelled |
git config --global user.email "[email protected]" | First-time setup — set your email for commit history |
git clone <repo-url> | Download a GitHub repo to your computer for the first time |
git status | See which files you have changed since your last commit |
git add . | Stage all your changes, ready to commit |
git commit -m "describe your change" | Save a checkpoint with a short message explaining what you changed |
git push | Upload your commits to GitHub. Before dev deployment, obtain Requirements Agent use-case approval. Push ready code to main, then use the Developer Guide request form with the submitted PRD link from All requests. |
git pull | Download the latest changes from GitHub to your computer |
Where you write and run code
| Word | What it actually is |
|---|---|
| Terminal | A text window where you type commands to your computer. Also called the command line or CLI |
| Editor / IDE | The program where you write and edit code. VS Code is one example |
Commands you will type
| Command | What it does |
|---|---|
pwd | Show which folder you are currently in |
ls (Mac) / dir (Windows) | List all files and folders in the current folder |
cd folder-name | Move into a folder (replace folder-name with your folder's name) |
cd .. | Move up one level, out of the current folder |
This platform
| Word | What it actually is |
|---|---|
| Deploy | Publish your code so it runs on the internet as a live webpage |
| call_llm | A helper provided by this platform that lets your app talk to an AI model (powered by AWS Bedrock) without any setup |
AI helpers
| Word | What it actually is |
|---|---|
| Microsoft 365 Copilot | The AI chat assistant in Teams, Word, and the Microsoft 365 home. ERM provides it |
| GitHub Copilot | A different Copilot that runs inside a code editor and writes code with you. Separate license |
Deploy prerequisites
These are not terms but hard rules — get any of them wrong and your deploy will fail. Keep them in mind before your first push.
| Rule | What it means |
|---|---|
| Repo name prefix | New Streamlit and custom React/FastAPI apps use ai-marketplace-<name>. Existing capability-tool-* repos are legacy-supported, but new apps should not use that prefix. |
| IT Service Desk registration | Before your tool can deploy to prod, IT must register your tool name in Azure AD. Raise a Service Desk ticket — see the Developer Guide deploy flow for the ticket body |
Streamlit basics
These are the Streamlit building blocks, copyable patterns, and rerun rules most tools need. Use this section if you are building with Streamlit. For non-Streamlit React/FastAPI apps, use the Developer Guide and the custom app template instead.
The template repo also includes runnable examples in examples/cheat_sheet.py.
Components
| I need to... | Use this |
|---|---|
| Upload a file | st.file_uploader("Label", type=["csv", "xlsx", "pdf"]) |
| Show a table | st.dataframe(df) for an interactive table, or st.table(df) for a static table |
| Let the user download a result | st.download_button("Download", data=..., file_name="out.xlsx") |
Show more Streamlit components
| I need to... | Use this |
|---|---|
| Show a running spinner | with st.spinner("Working..."): |
| Show multi-step progress | with st.status("Running...", expanded=True) as s: |
| Side-by-side layout | col1, col2 = st.columns(2) |
| Tabbed sections | tab1, tab2 = st.tabs(["A", "B"]) |
| Hide details until expanded | with st.expander("Show more"): |
| Batched form submission | with st.form("key"): plus st.form_submit_button() |
| Persist values across reruns | st.session_state["key"] = value |
| Cache an expensive call | @st.cache_data for data, or @st.cache_resource for models and clients |
Common patterns
Short snippets you can copy into app.py. Each is complete and runnable.
Upload, parse, preview
import streamlit as st
import pandas as pd
uploaded = st.file_uploader("CSV or Excel", type=["csv", "xlsx"])
if uploaded:
df = pd.read_csv(uploaded) if uploaded.name.endswith(".csv") else pd.read_excel(uploaded)
st.dataframe(df.head(20))
st.metric("Rows", len(df))
Call the LLM with a spinner
import streamlit as st
from platform_config import call_llm
text = st.text_area("Paste your text")
if st.button("Summarize") and text:
with st.spinner("Thinking..."):
result = call_llm(f"Summarize in 3 bullets:\n\n{text}")
st.write(result)
Export a DataFrame to Excel
import io
import pandas as pd
import streamlit as st
df = pd.DataFrame({"name": ["Alice", "Bob"], "score": [92, 78]})
buf = io.BytesIO()
with pd.ExcelWriter(buf, engine="openpyxl") as writer:
df.to_excel(writer, index=False, sheet_name="Results")
st.download_button(
"Download Excel",
data=buf.getvalue(),
file_name="results.xlsx",
mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
)
Multi-step progress with st.status
import streamlit as st
with st.status("Processing...", expanded=True) as status:
st.write("Step 1: Loading data")
# ... work ...
st.write("Step 2: Running analysis")
# ... work ...
st.write("Step 3: Building report")
status.update(label="Done", state="complete")
State and reruns
Streamlit re-runs your entire script top-to-bottom on every interaction. If you need a value to survive, put it in st.session_state.
| Situation | Use |
|---|---|
| A variable must survive between clicks | st.session_state["my_key"] |
| An expensive computation should only run once per input | @st.cache_data |
| A client or model should load only once per session | @st.cache_resource |
| You want to force a rerun from code | st.rerun() |
| You need user input before doing work | Gate the logic behind if st.button(...) or st.form_submit_button() |
AI-based learning
If the cheat sheet above is not enough and you want an AI to walk you through something hands-on — installing Python, using a terminal, decoding an error — the prompts below will do the heavy lifting. Each one is a ready-to-paste question scoped to a single task.
Where to paste them
Microsoft 365 Copilot is the easiest starting point if ERM has already given you a license. Open it from the Copilot icon in Microsoft Teams, or go to m365.cloud.microsoft/chat and sign in with your ERM email.
If you do not have a Copilot license yet, raise a ticket with the IT Service Desk to request Microsoft 365 Copilot access. If you already use a different AI chat tool, that works too — the prompts are plain text and any modern AI chat can answer them. Do not paste client confidential or sensitive data into AI tools unless the use case has been assessed; see the AI security and client data guidance.
How to use a prompt
Copy the whole prompt (including the blank lines), paste it into a new chat, and have a short conversation. When you are done with one topic, start a fresh chat for the next so the AI starts clean. These are suggestions — skip any you already know how to do.
1. Install Python, Poetry, Git, and a code editor
Use when: you are setting up your computer from scratch.
2. Use a terminal for the first time
Use when: you need to run commands on your computer but have never opened a terminal.
3. Connect Git to GitHub and clone a repo
Use when: you have Git installed and need to download a project from GitHub to your computer.
4. Run a Streamlit app on your own computer
Use when: you have a folder with a Python file called app.py and want to see it as a web page on your own machine.
5. Commit and push changes back to GitHub
Use when: you edited a file in a cloned repo and need to upload the changes back to GitHub.
6. Decode a Python error message
Use when: you ran a command and got a wall of red text you do not understand.
7. Show me the Streamlit widgets I can use
Use when: you see widgets like st.text_input in code and want to know what options exist.
When you are ready
You are ready to go back to the Developer Guide when you have done these things at least once:
| Can you do this? |
|---|
| Install Python, Poetry, Git, and a code editor on your computer |
| Open a terminal and run a basic command |
| Clone a GitHub repo to your computer |
Run a Streamlit app locally with poetry run streamlit run app.py, or run the custom React/FastAPI template's backend and frontend from its README. |
| Commit a change and push it to GitHub |
If any of those are still fuzzy, check the cheat sheet above, try the relevant AI prompt, or ask someone who has done it before. Once you can do all five, head back to the Developer Guide — you will recognise almost everything it asks you to do.