From 0760db03f81eb794cbcea84b1da27f7d970cbf3a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Thu, 6 Jul 2023 16:15:41 +0200
Subject: [PATCH 1/8] Perf: Retry for a long time, and also when throttling

---
 src/conversation-ui/src/Utils.jsx | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/src/conversation-ui/src/Utils.jsx b/src/conversation-ui/src/Utils.jsx
index ad9ef4d2..943ec94d 100644
--- a/src/conversation-ui/src/Utils.jsx
+++ b/src/conversation-ui/src/Utils.jsx
@@ -7,10 +7,12 @@ const client = axios.create({
   baseURL: 'http://127.0.0.1:8081',
 });
 axiosRetry(client, {
-  retries: 3,
-  retryCondition: axiosRetry.isNetworkOrIdempotentRequestError,
+  retries: 12,
   retryDelay: axiosRetry.exponentialDelay,
   shouldResetTimeout: true,
+  retryCondition: (error) => {
+    return axiosRetry.isNetworkOrIdempotentRequestError(error) || error.response.status == 429;
+  }
 });
 
 const header = (enabled) => {

From 9568fccf3d8a689db86cbf197f66b1001d9096cb Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Thu, 6 Jul 2023 16:15:55 +0200
Subject: [PATCH 2/8] Quality: Ignore ViteJS tmp file

---
 src/conversation-ui/.gitignore | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/src/conversation-ui/.gitignore b/src/conversation-ui/.gitignore
index 3d3bb879..5c78e512 100644
--- a/src/conversation-ui/.gitignore
+++ b/src/conversation-ui/.gitignore
@@ -1,3 +1,6 @@
+# Vite.js
+.vite/
+
 # Created by https://www.toptal.com/developers/gitignore/api/node,yarn,react,venv
 # Edit at https://www.toptal.com/developers/gitignore?templates=node,yarn,react,venv
 

From 3ba03f01108a9d5d63d2dede02833fdba39bcd76 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Thu, 6 Jul 2023 16:40:14 +0200
Subject: [PATCH 3/8] Perf: Rate-limit API by auth header

Allows to not throttle users using the same IP with different accounts.
---
 .../private-gpt/templates/conversation-api-ingress.yaml    | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/cicd/helm/private-gpt/templates/conversation-api-ingress.yaml b/cicd/helm/private-gpt/templates/conversation-api-ingress.yaml
index 379aa95c..d1b48a98 100644
--- a/cicd/helm/private-gpt/templates/conversation-api-ingress.yaml
+++ b/cicd/helm/private-gpt/templates/conversation-api-ingress.yaml
@@ -17,7 +17,7 @@ spec:
       middlewares:
         - name: {{ include "private-gpt.fullname" . }}-conversation-api-prefix
         - name: {{ include "private-gpt.fullname" . }}-conversation-api-security
-        - name: {{ include "private-gpt.fullname" . }}-conversation-api-ratelimit
+        - name: {{ include "private-gpt.fullname" . }}-conversation-api-ratelimit-auth
         - name: {{ include "private-gpt.fullname" . }}-conversation-api-compress
   tls:
     {{- toYaml .Values.ingress.tls | nindent 4 }}
@@ -49,7 +49,7 @@ spec:
 apiVersion: traefik.containo.us/v1alpha1
 kind: Middleware
 metadata:
-  name: {{ include "private-gpt.fullname" . }}-conversation-api-ratelimit
+  name: {{ include "private-gpt.fullname" . }}-conversation-api-ratelimit-auth
   labels:
     {{- include "private-gpt.labels" . | nindent 4 }}
     app.kubernetes.io/component: conversation-api
@@ -57,6 +57,9 @@ spec:
   rateLimit:
     average: 1
     burst: 5
+    period: 1s
+    sourceCriterion:
+      requestHeaderName: Authorization
 ---
 apiVersion: traefik.containo.us/v1alpha1
 kind: Middleware

From ef8697a6896e1f735efb7789b95de7a0cd8f150c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Thu, 6 Jul 2023 16:55:09 +0200
Subject: [PATCH 4/8] Feat: Make header title clickable

---
 src/conversation-ui/src/Header.jsx |  6 ++++--
 src/conversation-ui/src/main.scss  | 21 +++++++++++++--------
 2 files changed, 17 insertions(+), 10 deletions(-)

diff --git a/src/conversation-ui/src/Header.jsx b/src/conversation-ui/src/Header.jsx
index 3445a5e0..fe73164a 100644
--- a/src/conversation-ui/src/Header.jsx
+++ b/src/conversation-ui/src/Header.jsx
@@ -3,7 +3,7 @@ import { header } from "./Utils";
 import { ThemeContext, ConversationContext } from "./App";
 import { useAuth } from "oidc-react";
 import { useContext } from "react";
-import { useNavigate } from "react-router-dom";
+import { useNavigate, Link } from "react-router-dom";
 import Button from "./Button";
 import Conversations from "./Conversations";
 
@@ -18,7 +18,9 @@ function Header() {
   return (
     <div className="header">
       <div className="header__top">
-        <h1>🔒 Private GPT</h1>
+        <Link to="/" className="a--unstyled">
+          <h1>🔒 Private GPT</h1>
+        </Link>
         <Button
           className="header__top__toggle"
           emoji="="
diff --git a/src/conversation-ui/src/main.scss b/src/conversation-ui/src/main.scss
index 088636e0..c662f460 100644
--- a/src/conversation-ui/src/main.scss
+++ b/src/conversation-ui/src/main.scss
@@ -142,7 +142,7 @@ body {
   textarea,
   button,
   input,
-  a {
+  a:not(.a--unstyled) {
     &,
     &::placeholder,
     &:disabled,
@@ -198,18 +198,23 @@ body {
   }
 
   a {
-    color: var(--color-active);
+    color: inherit;
     text-decoration: none;
 
-    &:not(:disabled):not([disabled]) {
-      &:hover {
-        text-decoration: underline;
-      }
+    &:not(.a--unstyled) {
+      color: var(--color-active);
 
-      &:visited {
-        color: var(--color-visited);
+      &:not(:disabled):not([disabled]) {
+        &:hover {
+          text-decoration: underline;
+        }
+
+        &:visited {
+          color: var(--color-visited);
+        }
       }
     }
+
   }
 
   ol {

From e1f357c740c8cd664b2dedc01b4dc3a2f1693490 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Thu, 6 Jul 2023 17:50:56 +0200
Subject: [PATCH 5/8] Feat: Add usage examples in first page

---
 src/conversation-ui/src/Conversation.jsx  | 26 ++++++++++++++++++++++
 src/conversation-ui/src/conversation.scss | 27 ++++++++++++++++++++++-
 2 files changed, 52 insertions(+), 1 deletion(-)

diff --git a/src/conversation-ui/src/Conversation.jsx b/src/conversation-ui/src/Conversation.jsx
index 40bd960e..374aba96 100644
--- a/src/conversation-ui/src/Conversation.jsx
+++ b/src/conversation-ui/src/Conversation.jsx
@@ -255,6 +255,32 @@ function Conversation() {
         {conversation.messages.length == 0 && (
           <div className="conversation__messages__empty">
             <big>🔒 Private GPT</big>
+            <div className="conversation__messages__empty__doc">
+              <div>
+                <h2>💡</h2>
+              </div>
+              <div>
+                <h2>Examples</h2>
+                <p>Generating content that is tailored to specific audiences or personas.</p>
+                <p>Help on technical language and jargon in specific industries (such as finance or healthcare).</p>
+              </div>
+              <div>
+                <h2>⚡️</h2>
+              </div>
+              <div>
+                <h2>Capabilities</h2>
+                <p>Analyzing large amounts of qualitative data (such as news articles or earnings calls) to inform investment decisions.</p>
+                <p>Generating summaries of court cases or contracts to save time on manual review.</p>
+              </div>
+              <div>
+                <h2>⚠️</h2>
+              </div>
+              <div>
+                <h2>Limitations</h2>
+                <p>Ensuring that generated content adheres to applicable regulations in a given industry or region.</p>
+                <p>Mitigating any risks associated with automated content creation (such as reputational harm or inadvertent bias).</p>
+              </div>
+            </div>
             {!auth.userData && (
               <Button
                 onClick={() => auth.signIn()}
diff --git a/src/conversation-ui/src/conversation.scss b/src/conversation-ui/src/conversation.scss
index 95a9e4c6..4af8f657 100644
--- a/src/conversation-ui/src/conversation.scss
+++ b/src/conversation-ui/src/conversation.scss
@@ -19,9 +19,9 @@
     }
 
     .conversation__messages__empty {
+      align-items: center;
       display: flex;
       flex-direction: column;
-      align-items: center;
 
       // Center the content vertically and horizontally
       align-self: center;
@@ -31,6 +31,31 @@
         margin-bottom: 1rem;
       }
     }
+
+    .conversation__messages__empty__doc {
+      background-color: var(--background-muted);
+      border-radius: var(--radius);
+      display: grid;
+      grid-gap: 1rem;
+      padding: 2rem;
+      place-items: flex-start;
+
+      // Small width
+      @media (max-width: 84rem) {
+        grid-template-columns: auto auto;
+      }
+
+      // Full width
+      @media (min-width: 84rem) {
+        grid-template-columns: auto auto auto auto auto auto;
+      }
+
+      div {
+        > *:not(:last-child) {
+          margin-bottom: 0.5rem;
+        }
+      }
+    }
   }
 
   .conversation__input {

From fc0cda6a3d19ee269847567ab1e43afab3e56c59 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Thu, 6 Jul 2023 17:51:09 +0200
Subject: [PATCH 6/8] Quality: CSS documentation

---
 src/conversation-ui/src/button.scss | 1 +
 src/conversation-ui/src/main.scss   | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/src/conversation-ui/src/button.scss b/src/conversation-ui/src/button.scss
index 6be91631..7028f4af 100644
--- a/src/conversation-ui/src/button.scss
+++ b/src/conversation-ui/src/button.scss
@@ -10,6 +10,7 @@
     background-color: var(--background-active);
   }
 
+  // Small width
   @media (max-width: 56rem) {
     .button__text {
       display: none;
diff --git a/src/conversation-ui/src/main.scss b/src/conversation-ui/src/main.scss
index c662f460..d394bd00 100644
--- a/src/conversation-ui/src/main.scss
+++ b/src/conversation-ui/src/main.scss
@@ -12,12 +12,14 @@ html {
   --radius: 1.5em; // Not relative, radius should be the same everywhere
   --search-padding-v: 1rem;
 
+  // Small width
   @media (max-width: 56rem) {
     --conversation-padding-h: 1rem;
     --header-padding-h: 1rem;
     --search-padding-h: 1rem;
   }
 
+  // Full width
   @media (min-width: 56rem) {
     --conversation-padding-h: 2rem;
     --header-padding-h: 2rem;

From e2d69b8fe6d1de65b707bc20f0263b8e73fe4362 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Fri, 7 Jul 2023 19:03:15 +0200
Subject: [PATCH 7/8] Feat: Add plug-and-play storage system and add Azure
 Cosmos DB store

---
 README.md                                   |  14 +++
 src/conversation-api/main.py                |  99 +++++++++++------
 src/conversation-api/models/message.py      |  11 +-
 src/conversation-api/models/prompt.py       |   2 +-
 src/conversation-api/persistence/cosmos.py  | 115 ++++++++++++++++++++
 src/conversation-api/persistence/isearch.py |  11 +-
 src/conversation-api/persistence/istore.py  |  12 +-
 src/conversation-api/persistence/istream.py |   9 +-
 src/conversation-api/persistence/qdrant.py  |  18 ++-
 src/conversation-api/persistence/redis.py   |  69 ++++++------
 src/conversation-api/requirements.txt       |   1 +
 src/conversation-api/utils.py               |  13 ++-
 12 files changed, 280 insertions(+), 94 deletions(-)
 create mode 100644 src/conversation-api/persistence/cosmos.py

diff --git a/README.md b/README.md
index e1352fa0..90adc8aa 100644
--- a/README.md
+++ b/README.md
@@ -9,6 +9,7 @@ Includes:
 - Dead simple interface
 - Manage users effortlessly with OpenID Connect
 - More than 150 tones and personalities (accountant, advisor, debater, excel sheet, instructor, logistician, etc.) to better help employees in their specific daily tasks
+- Plug and play with any storage system, including [Azure Cosmos DB](https://learn.microsoft.com/en-us/azure/cosmos-db/), [Redis](https://github.com/redis/redis) and [Qdrant](https://github.com/qdrant/qdrant).
 - Possibility to send temporary messages, for confidentiality
 - Salable system based on stateless APIs, cache, progressive web app and events
 - Search engine for conversations, based on semantic similarity and AI embeddings
@@ -25,6 +26,14 @@ Create a local configuration file, a file named `config.toml` at the root of the
 ```toml
 # config.toml
 # Values are for example only, you should change them
+[persistence]
+# Enum: "qdrant"
+search = "qdrant"
+# Enum: "redis", "cosmos"
+store = "cosmos"
+# Enum: "redis"
+stream = "redis"
+
 [openai]
 ada_deploy_id = "ada"
 ada_max_tokens = 2049
@@ -53,6 +62,11 @@ host = "localhost"
 [redis]
 db = 0
 host = "localhost"
+
+[cosmos]
+# Containers "conversation" (/user_id), "message" (/conversation_id) and "user" (/dummy) must exist
+url = "https://private-gpt.documents.azure.com:443"
+database = "private-gpt"
 ```
 
 Now, you can either run the application as container or with live reload. For development, it is recommended to use live reload. For demo, it is recommended to use the container.
diff --git a/src/conversation-api/main.py b/src/conversation-api/main.py
index e7c81e33..48ff6bcc 100644
--- a/src/conversation-api/main.py
+++ b/src/conversation-api/main.py
@@ -15,17 +15,14 @@
 from fastapi import FastAPI, HTTPException, status, Request, Depends
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
-from models.conversation import (
-    GetConversationModel,
-    ListConversationsModel,
-    StoredConversationModel,
-)
-from models.message import MessageModel, MessageRole
+from models.conversation import (GetConversationModel, ListConversationsModel, StoredConversationModel)
+from models.message import MessageModel, MessageRole, StoredMessageModel
 from models.prompt import StoredPromptModel, ListPromptsModel
 from models.search import SearchModel
 from models.user import UserModel
-from persistence.qdrant import QdrantSearch
-from persistence.redis import RedisStore, RedisStream, STREAM_STOPWORD
+from persistence.isearch import SearchImplementation
+from persistence.istore import StoreImplementation
+from persistence.istream import StreamImplementation
 from sse_starlette.sse import EventSourceResponse
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 from typing import Annotated, Dict, List, Optional
@@ -45,12 +42,35 @@
 logger = build_logger(__name__)
 
 ###
-# Init Redis
+# Init persistence
 ###
-
-store = RedisStore()
-stream = RedisStream()
-index = QdrantSearch(store)
+store_impl = get_config("persistence", "store", StoreImplementation, required=True)
+if store_impl == StoreImplementation.COSMOS:
+    logger.info("Using CosmosDB store")
+    from persistence.cosmos import CosmosStore
+    store = CosmosStore()
+elif store_impl == StoreImplementation.REDIS:
+    logger.info("Using Redis store")
+    from persistence.redis import RedisStore
+    store = RedisStore()
+else:
+    raise ValueError(f"Unknown store implementation: {store_impl}")
+
+search_impl = get_config("persistence", "search", SearchImplementation, required=True)
+if search_impl == SearchImplementation.QDRANT:
+    logger.info("Using Qdrant search")
+    from persistence.qdrant import QdrantSearch
+    index = QdrantSearch(store)
+else:
+    raise ValueError(f"Unknown search implementation: {search_impl}")
+
+stream_impl = get_config("persistence", "stream", StreamImplementation, required=True)
+if stream_impl == StreamImplementation.REDIS:
+    logger.info("Using Redis stream")
+    from persistence.redis import RedisStream, STREAM_STOPWORD
+    stream = RedisStream()
+else:
+    raise ValueError(f"Unknown stream implementation: {stream_impl}")
 
 ###
 # Init OpenAI
@@ -299,15 +319,6 @@ async def message_post(
             detail="Message is moderated",
         )
 
-    message = MessageModel(
-        content=content,
-        created_at=datetime.now(),
-        id=uuid4(),
-        role=MessageRole.USER,
-        secret=secret,
-        token=uuid4(),
-    )
-
     if conversation_id:
         # Validate API schema
         if prompt_id:
@@ -326,12 +337,24 @@ async def message_post(
                 detail="Conversation not found",
             )
 
+        # Build message
+        message = StoredMessageModel(
+            content=content,
+            conversation_id=conversation_id,
+            created_at=datetime.now(),
+            id=uuid4(),
+            role=MessageRole.USER,
+            secret=secret,
+            token=uuid4(),
+        )
+
         # Validate message length
         tokens_nb = oai_tokens_nb(
             message.content
-            + "".join([m.content for m in store.message_list(conversation_id)]),
+            + "".join([m.content for m in store.message_list(message.conversation_id)]),
             OAI_GPT_MODEL,
         )
+
         logger.debug(f"{tokens_nb} tokens in the conversation")
         if tokens_nb > OAI_GPT_MAX_TOKENS:
             logger.info(f"Message ({tokens_nb}) too long for conversation")
@@ -341,8 +364,8 @@ async def message_post(
             )
 
         # Update conversation
-        store.message_set(message, conversation_id)
-        index.message_index(message, conversation_id, current_user.id)
+        store.message_set(message)
+        index.message_index(message, current_user.id)
         conversation = store.conversation_get(conversation_id, current_user.id)
         if not conversation:
             logger.warn("ACID error: conversation not found after testing existence")
@@ -359,7 +382,7 @@ async def message_post(
             )
 
         # Validate message length
-        tokens_nb = oai_tokens_nb(message.content, OAI_GPT_MODEL)
+        tokens_nb = oai_tokens_nb(content, OAI_GPT_MODEL)
         logger.debug(f"{tokens_nb} tokens in the conversation")
         if tokens_nb > OAI_GPT_MAX_TOKENS:
             logger.info(f"Message ({tokens_nb}) too long for conversation")
@@ -368,7 +391,7 @@ async def message_post(
                 detail="Conversation history is too long",
             )
 
-        # Create a new conversation
+        # Build conversation
         conversation = StoredConversationModel(
             created_at=datetime.now(),
             id=uuid4(),
@@ -376,8 +399,19 @@ async def message_post(
             user_id=current_user.id,
         )
         store.conversation_set(conversation)
-        store.message_set(message, conversation.id)
-        index.message_index(message, conversation.id, current_user.id)
+
+        # Build message
+        message = StoredMessageModel(
+            content=content,
+            conversation_id=conversation.id,
+            created_at=datetime.now(),
+            id=uuid4(),
+            role=MessageRole.USER,
+            secret=secret,
+            token=uuid4(),
+        )
+        store.message_set(message)
+        index.message_index(message, current_user.id)
 
     messages = store.message_list(conversation.id)
 
@@ -486,15 +520,16 @@ def completion_from_conversation(
             content_full += content
 
     # First, store the updated conversation in Redis
-    res_message = MessageModel(
+    res_message = StoredMessageModel(
         content=content_full,
+        conversation_id=conversation.id,
         created_at=datetime.now(),
         id=uuid4(),
         role=MessageRole.ASSISTANT,
         secret=last_message.secret,
     )
-    store.message_set(res_message, conversation.id)
-    index.message_index(res_message, conversation.id, current_user.id)
+    store.message_set(res_message)
+    index.message_index(res_message, current_user.id)
 
     # Then, send the end of stream message
     stream.push(STREAM_STOPWORD, last_message.token)
diff --git a/src/conversation-api/models/message.py b/src/conversation-api/models/message.py
index 0682f0bd..ebb73411 100644
--- a/src/conversation-api/models/message.py
+++ b/src/conversation-api/models/message.py
@@ -1,7 +1,7 @@
 from datetime import datetime
 from enum import Enum
-from typing import List, Optional
 from pydantic import BaseModel
+from typing import Optional
 from uuid import UUID
 
 
@@ -20,7 +20,16 @@ class MessageModel(BaseModel):
     token: Optional[UUID] = None
 
 
+class StoredMessageModel(MessageModel):
+    conversation_id: UUID
+
+
 class IndexMessageModel(BaseModel):
+    """
+    Storing the message in a separate collection allows us to query for messages. It does not contain the message content, but only the metadata required to query for the message.
+
+    The absence of content and created_at is intentional. We don't want to store PII in the index. As this, we are not forced to apply a TTL to the index nor secure too much the DB.
+    """
     conversation_id: UUID
     id: UUID
     user_id: UUID
diff --git a/src/conversation-api/models/prompt.py b/src/conversation-api/models/prompt.py
index 9ccad94c..d2016f4e 100644
--- a/src/conversation-api/models/prompt.py
+++ b/src/conversation-api/models/prompt.py
@@ -1,5 +1,5 @@
-from typing import List
 from pydantic import BaseModel
+from typing import List
 from uuid import UUID
 
 
diff --git a/src/conversation-api/persistence/cosmos.py b/src/conversation-api/persistence/cosmos.py
new file mode 100644
index 00000000..a1eca0a9
--- /dev/null
+++ b/src/conversation-api/persistence/cosmos.py
@@ -0,0 +1,115 @@
+# Import utils
+from utils import build_logger, get_config
+
+# Import misc
+from .istore import IStore
+from azure.cosmos import CosmosClient, PartitionKey, ConsistencyLevel
+from azure.cosmos.database import DatabaseProxy
+from azure.cosmos.exceptions import CosmosHttpResponseError, CosmosResourceExistsError
+from azure.identity import DefaultAzureCredential
+from datetime import datetime
+from models.conversation import StoredConversationModel, StoredConversationModel
+from models.message import MessageModel, IndexMessageModel, StoredMessageModel
+from models.user import UserModel
+from typing import (Any, Dict, List, Union)
+from uuid import UUID
+
+
+logger = build_logger(__name__)
+SECRET_TTL_SECS = 60 * 60 * 24  # 1 day
+
+# Configuration
+CONVERSATION_PREFIX = "conversation"
+DB_URL = get_config("cosmos", "url", str, required=True)
+DB_NAME = get_config("cosmos", "database", str, required=True)
+
+# Cosmos DB Client
+credential = DefaultAzureCredential()
+client = CosmosClient(url=DB_URL, credential=credential, consistency_level=ConsistencyLevel.Session)
+database = client.get_database_client(DB_NAME)
+conversation_client = database.get_container_client("conversation")
+message_client = database.get_container_client("message")
+user_client = database.get_container_client("user")
+logger.info(f'Connected to Cosmos DB at "{DB_URL}"')
+
+
+class CosmosStore(IStore):
+    def user_get(self, user_external_id: str) -> Union[UserModel, None]:
+        query = f"SELECT * FROM c WHERE c.external_id = '{user_external_id}'"
+        items = user_client.query_items(query=query, partition_key="dummy")
+        try:
+            raw = next(items)
+            return UserModel(**raw)
+        except StopIteration:
+            return None
+
+    def user_set(self, user: UserModel) -> None:
+        user_client.upsert_item(body={
+            **self._sanitize_before_insert(user.dict()),
+            "dummy": "dummy",
+        })
+
+    def conversation_get(
+        self, conversation_id: UUID, user_id: UUID
+    ) -> Union[StoredConversationModel, None]:
+        try:
+            raw = conversation_client.read_item(
+                item=str(conversation_id), partition_key=str(user_id)
+            )
+            return StoredConversationModel(**raw)
+        except CosmosHttpResponseError:
+            return None
+
+    def conversation_exists(self, conversation_id: UUID, user_id: UUID) -> bool:
+        return self.conversation_get(conversation_id, user_id) is not None
+
+    def conversation_set(self, conversation: StoredConversationModel) -> None:
+        conversation_client.upsert_item(body=self._sanitize_before_insert(conversation.dict()))
+
+    def conversation_list(self, user_id: UUID) -> List[StoredConversationModel]:
+        query = f"SELECT * FROM c WHERE c.user_id = '{user_id}'"
+        items = conversation_client.query_items(query=query, enable_cross_partition_query=True)
+        return [StoredConversationModel(**item) for item in items]
+
+    def message_get(
+        self, message_id: UUID, conversation_id: UUID
+    ) -> Union[MessageModel, None]:
+        try:
+            raw = message_client.read_item(item=str(message_id), partition_key=str(conversation_id))
+            return MessageModel(**raw)
+        except CosmosHttpResponseError:
+            return None
+
+    def message_get_index(
+        self, message_indexs: List[IndexMessageModel]
+    ) -> List[MessageModel]:
+        messages = []
+        for message_index in message_indexs:
+            try:
+                raw = message_client.read_item(
+                    item=str(message_index.id), partition_key=str(message_index.conversation_id)
+                )
+                messages.append(MessageModel(**raw))
+            except CosmosHttpResponseError:
+                pass
+        return messages
+
+    def message_set(self, message: StoredMessageModel) -> None:
+        expiry = SECRET_TTL_SECS if message.secret else None
+        message_client.upsert_item(body={
+            **self._sanitize_before_insert(message.dict()),
+            "_ts": expiry, # TTL in seconds
+        })
+
+    def message_list(self, conversation_id: UUID) -> List[MessageModel]:
+        query = f"SELECT * FROM c WHERE c.conversation_id = '{conversation_id}'"
+        items = message_client.query_items(query=query, enable_cross_partition_query=True)
+        return [MessageModel(**item) for item in items]
+
+    def _sanitize_before_insert(self, item: dict) -> Dict[str, Union[str, int, float, bool]]:
+        for key, value in item.items():
+            if isinstance(value, UUID):
+                item[key] = str(value)
+            elif isinstance(value, datetime):
+                item[key] = value.isoformat()
+        return item
diff --git a/src/conversation-api/persistence/isearch.py b/src/conversation-api/persistence/isearch.py
index b6876731..f399ae04 100644
--- a/src/conversation-api/persistence/isearch.py
+++ b/src/conversation-api/persistence/isearch.py
@@ -1,10 +1,15 @@
 from .istore import IStore
 from abc import ABC, abstractmethod
-from models.message import MessageModel
+from enum import Enum
+from models.message import MessageModel, StoredMessageModel
 from models.search import SearchModel
 from uuid import UUID
 
 
+class SearchImplementation(str, Enum):
+    QDRANT = "qdrant"
+
+
 class ISearch(ABC):
     def __init__(self, store: IStore):
         self.store = store
@@ -14,7 +19,5 @@ def message_search(self, query: str, user_id: UUID) -> SearchModel[MessageModel]
         pass
 
     @abstractmethod
-    def message_index(
-        self, message: MessageModel, conversation_id: UUID, user_id: UUID
-    ) -> None:
+    def message_index(self, message: StoredMessageModel, user_id: UUID) -> None:
         pass
diff --git a/src/conversation-api/persistence/istore.py b/src/conversation-api/persistence/istore.py
index baa25423..0433db74 100644
--- a/src/conversation-api/persistence/istore.py
+++ b/src/conversation-api/persistence/istore.py
@@ -1,11 +1,17 @@
+from abc import ABC, abstractmethod
+from enum import Enum
 from models.conversation import GetConversationModel, StoredConversationModel
-from models.message import MessageModel, IndexMessageModel
+from models.message import MessageModel, IndexMessageModel, StoredMessageModel
 from models.user import UserModel
-from abc import ABC, abstractmethod
 from typing import List, Union
 from uuid import UUID
 
 
+class StoreImplementation(str, Enum):
+    COSMOS = "cosmos"
+    REDIS = "redis"
+
+
 class IStore(ABC):
     @abstractmethod
     def user_get(self, user_external_id: str) -> Union[UserModel, None]:
@@ -46,7 +52,7 @@ def message_get(
         pass
 
     @abstractmethod
-    def message_set(self, message: MessageModel, conversation_id: UUID) -> None:
+    def message_set(self, message: StoredMessageModel) -> None:
         pass
 
     @abstractmethod
diff --git a/src/conversation-api/persistence/istream.py b/src/conversation-api/persistence/istream.py
index 1b93f4be..a02a3c24 100644
--- a/src/conversation-api/persistence/istream.py
+++ b/src/conversation-api/persistence/istream.py
@@ -1,8 +1,13 @@
 from abc import ABC, abstractmethod
-from typing import AsyncGenerator, Callable, Literal
+from enum import Enum
+from typing import AsyncGenerator, Callable, Literal, Any
 from uuid import UUID
 
 
+class StreamImplementation(str, Enum):
+    REDIS = "redis"
+
+
 class IStream(ABC):
     @abstractmethod
     def push(self, content: str, token: UUID) -> None:
@@ -11,7 +16,7 @@ def push(self, content: str, token: UUID) -> None:
     @abstractmethod
     async def get(
         self, token: UUID, loop_func: Callable[[], bool]
-    ) -> AsyncGenerator[any | Literal["STOP"], None]:
+    ) -> AsyncGenerator[Any | Literal["STOP"], None]:
         pass
 
     @abstractmethod
diff --git a/src/conversation-api/persistence/qdrant.py b/src/conversation-api/persistence/qdrant.py
index 21716608..a70c768c 100644
--- a/src/conversation-api/persistence/qdrant.py
+++ b/src/conversation-api/persistence/qdrant.py
@@ -5,7 +5,7 @@
 from .isearch import ISearch
 from .istore import IStore
 from datetime import datetime
-from models.message import MessageModel, IndexMessageModel
+from models.message import MessageModel, IndexMessageModel, StoredMessageModel
 from models.search import SearchModel, SearchStatsModel, SearchAnswerModel
 from qdrant_client import QdrantClient
 from tenacity import retry, stop_after_attempt, wait_random_exponential
@@ -62,11 +62,7 @@ def message_search(self, q: str, user_id: UUID) -> SearchModel[MessageModel]:
         vector = self._vector_from_text(
             textwrap.dedent(
                 f"""
-                Today, we are the {datetime.now()}.
-
-                QUERY START
-                {q}
-                QUERY END
+                Today, we are the {datetime.now()}. {q.capitalize()}
             """
             ),
             user_id,
@@ -108,21 +104,23 @@ def message_search(self, q: str, user_id: UUID) -> SearchModel[MessageModel]:
         )
 
     def message_index(
-        self, message: MessageModel, conversation_id: UUID, user_id: UUID
+        self, message: StoredMessageModel, user_id: UUID
     ) -> None:
         logger.debug(f"Indexing message: {message.id}")
         self._loop.run_in_executor(
-            None, lambda: self._index_worker(message, conversation_id, user_id)
+            None, lambda: self._index_worker(message, user_id)
         )
 
     def _index_worker(
-        self, message: MessageModel, conversation_id: UUID, user_id: UUID
+        self, message: StoredMessageModel, user_id: UUID
     ) -> None:
         logger.debug(f"Starting indexing worker for message: {message.id}")
 
         vector = self._vector_from_text(message.content, user_id)
         index = IndexMessageModel(
-            conversation_id=conversation_id, id=message.id, user_id=user_id
+            conversation_id=message.conversation_id,
+            id=message.id,
+            user_id=user_id,
         )
 
         client.upsert(
diff --git a/src/conversation-api/persistence/redis.py b/src/conversation-api/persistence/redis.py
index c348dfef..4f2e0c74 100644
--- a/src/conversation-api/persistence/redis.py
+++ b/src/conversation-api/persistence/redis.py
@@ -5,34 +5,29 @@
 from .istore import IStore
 from .istream import IStream
 from models.conversation import StoredConversationModel, StoredConversationModel
-from models.message import MessageModel, IndexMessageModel
+from models.message import MessageModel, IndexMessageModel, StoredMessageModel
 from models.user import UserModel
 from redis import Redis
-from typing import (
-    Any,
-    AsyncGenerator,
-    Callable,
-    Awaitable,
-    List,
-    Literal,
-    Optional,
-    Union,
-)
+from typing import (Any, AsyncGenerator, Callable, Awaitable, List, Literal, Optional, Union)
 from uuid import UUID
 import asyncio
 
 
 logger = build_logger(__name__)
 SECRET_TTL_SECS = 60 * 60 * 24  # 1 day
+
+# Configuration
 CONVERSATION_PREFIX = "conversation"
 MESSAGE_PREFIX = "message"
-REDIS_HOST = get_config("redis", "host", str, required=True)
-REDIS_PORT = 6379
+DB_HOST = get_config("redis", "host", str, required=True)
+DB_PORT = 6379
 STREAM_PREFIX = "stream"
 STREAM_STOPWORD = "STOP"
 USER_PREFIX = "user"
-client = Redis(db=0, host=REDIS_HOST, port=REDIS_PORT)
-logger.info(f'Connected to Redis at "{REDIS_HOST}:{REDIS_PORT}"')
+
+# Redis client
+client = Redis(db=0, host=DB_HOST, port=DB_PORT)
+logger.info(f'Connected to Redis at "{DB_HOST}:{DB_PORT}"')
 
 
 class RedisStore(IStore):
@@ -53,26 +48,6 @@ def conversation_get(
             return None
         return StoredConversationModel.parse_raw(raw)
 
-    def message_get_index(
-        self, message_indexs: List[IndexMessageModel]
-    ) -> List[MessageModel]:
-        keys = [
-            self._message_cache_key(message_index.conversation_id, message_index.id)
-            for message_index in message_indexs
-        ]
-        raws = client.mget(keys)
-        if raws is None:
-            return []
-        messages = []
-        for raw in raws:
-            if raw is None:
-                continue
-            try:
-                messages.append(MessageModel.parse_raw(raw))
-            except Exception:
-                logger.warn("Error parsing message", exc_info=True)
-        return messages
-
     def conversation_exists(self, conversation_id: UUID, user_id: UUID) -> bool:
         return (
             client.exists(self._conversation_cache_key(user_id, conversation_id)) != 0
@@ -109,10 +84,30 @@ def message_get(
             return None
         return MessageModel.parse_raw(raw)
 
-    def message_set(self, message: MessageModel, conversation_id: UUID) -> None:
+    def message_get_index(
+        self, message_indexs: List[IndexMessageModel]
+    ) -> List[MessageModel]:
+        keys = [
+            self._message_cache_key(message_index.conversation_id, message_index.id)
+            for message_index in message_indexs
+        ]
+        raws = client.mget(keys)
+        if raws is None:
+            return []
+        messages = []
+        for raw in raws:
+            if raw is None:
+                continue
+            try:
+                messages.append(MessageModel.parse_raw(raw))
+            except Exception:
+                logger.warn("Error parsing message", exc_info=True)
+        return messages
+
+    def message_set(self, message: StoredMessageModel) -> None:
         expiry = SECRET_TTL_SECS if message.secret else None
         client.set(
-            self._message_cache_key(conversation_id, message.id),
+            self._message_cache_key(message.conversation_id, message.id),
             message.json(),
             ex=expiry,
         )
diff --git a/src/conversation-api/requirements.txt b/src/conversation-api/requirements.txt
index 91967585..13cfd603 100644
--- a/src/conversation-api/requirements.txt
+++ b/src/conversation-api/requirements.txt
@@ -1,5 +1,6 @@
 aiohttp==3.8.4
 azure-ai-contentsafety==1.0.0b1
+azure-cosmos==4.4.0
 azure-identity==1.13.0
 fastapi==0.95.2
 mmh3==4.0.0
diff --git a/src/conversation-api/utils.py b/src/conversation-api/utils.py
index a93f26ba..3cf0dd0b 100644
--- a/src/conversation-api/utils.py
+++ b/src/conversation-api/utils.py
@@ -65,14 +65,19 @@ def get_env(key: str, res_default: T) -> Union[str, T]:
 
     # Convert to res_type
     try:
-        if validate is bool:
+        if validate is bool: # bool
             res = res.strip().lower() == "true"
-        elif validate is int:
+        elif validate is int: # int
             res = int(res)
-        elif validate is float:
+        elif validate is float: # float
             res = float(res)
-        elif validate is UUID:
+        elif validate is UUID: # UUID
             res = UUID(res)
+        else: # Enum
+            try:
+                res = validate(res)
+            except Exception:
+                pass
     except Exception:
         raise ConfigNotFound(
             f'Cannot convert config "{section}/{key}" ({validate.__name__}), found "{res}" ({type(res).__name__})'

From 0a6c339c9ae99d5f02a82abc483825525391c09f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Cl=C3=A9mence=20Lesn=C3=A9?= <clemence@lesne.pro>
Date: Fri, 7 Jul 2023 19:03:26 +0200
Subject: [PATCH 8/8] Doc: Add Helm feature

---
 README.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/README.md b/README.md
index 90adc8aa..23706533 100644
--- a/README.md
+++ b/README.md
@@ -7,6 +7,7 @@ Includes:
 - Can be configured to use any Azure OpenAI completion API, including GPT-4
 - Dark theme for better readability
 - Dead simple interface
+- Deployable on any Kubernetes cluster, with its Helm chart
 - Manage users effortlessly with OpenID Connect
 - More than 150 tones and personalities (accountant, advisor, debater, excel sheet, instructor, logistician, etc.) to better help employees in their specific daily tasks
 - Plug and play with any storage system, including [Azure Cosmos DB](https://learn.microsoft.com/en-us/azure/cosmos-db/), [Redis](https://github.com/redis/redis) and [Qdrant](https://github.com/qdrant/qdrant).