diff --git a/website/docs/contributing/ADRs/ADRs.md b/website/docs/contributing/ADRs/ADRs.md
index d213083a37..ff307054b1 100644
--- a/website/docs/contributing/ADRs/ADRs.md
+++ b/website/docs/contributing/ADRs/ADRs.md
@@ -26,6 +26,7 @@ We are in the process of defining ADRs for the back end. At the time of writing
 * [POST/PUT API payload](./back-end/POST-PUT-api-payload.md)
 * [Specificity in database column references](./back-end/specificity-db-columns.md)
 * [Write model vs Read models](./back-end/write-model-vs-read-models.md)
+* [Frontend API Design](./back-end/frontend-api-design.md)
 
 ## Front-end ADRs
 
diff --git a/website/docs/contributing/ADRs/back-end/frontend-api-design.md b/website/docs/contributing/ADRs/back-end/frontend-api-design.md
new file mode 100644
index 0000000000..e2dec01124
--- /dev/null
+++ b/website/docs/contributing/ADRs/back-end/frontend-api-design.md
@@ -0,0 +1,34 @@
+---
+title: "ADR: Frontend API design"
+---
+
+## Background
+
+Previous version of the Frontend API (known as proxy API) had memory and I/O issues when a large number of tokens was used. 
+
+To better understand how it worked you need to know that Frontend API is Unleash Node SDK exposed over the API.
+Node SDK allows to plug in a custom repository so that it can fetch the toggles from a database instead of making remote HTTP calls. 
+
+Every time a new Frontend API token was used we created a new SDK client with its own proxy repository. Proxy repository was fetching
+the flags for a project and environment extracted from a token. 
+Every time there was a revision ID change (e.g. after some flag changed the status) the repository was updating itself. 
+Since every token had a dedicated repository with 10 tokens we were making 10 DB calls on each change. 
+What's more each repository kept its own copy of the flags. What it means in practice is that two different tokens with the same
+project and environments would store the same flags twice. 
+
+[Frontend API before](/img/frontend-api-before.png)
+
+## Decision
+
+To address these challenges, we came up with a new design:
+
+[Frontend API after](/img/frontend-api-after.png)
+
+We decided to swap ProxyRepository with a drop-in replacement FrontendApiRepository. FrontendApiRepository doesn't store any flags on its own but always filters
+the flags that we keep in a GlobalFrontendApiCache. The cache stores all the flags and updates on every revision ID change. 
+
+Consequences:
+* memory improvements: for a large number of tokens the memory footprint is reduced since we only store one copy of each flag in the cache and every repository
+filters the data that it needs for a given project and environment obtained from the token
+* I/O improvements: for a large number of tokens the number of DB calls is reduced to one per revision ID update since only the cache needs to be updated
+
diff --git a/website/static/img/frontend-api-after.png b/website/static/img/frontend-api-after.png
new file mode 100644
index 0000000000..8b730f4062
Binary files /dev/null and b/website/static/img/frontend-api-after.png differ
diff --git a/website/static/img/frontend-api-before.png b/website/static/img/frontend-api-before.png
new file mode 100644
index 0000000000..d01df44401
Binary files /dev/null and b/website/static/img/frontend-api-before.png differ