fix deployment

Dockerfile

@@ -21,5 +21,5 @@ RUN pip install -e ".[cpu]" "../sparql-llm"
 
 # https://github.com/tiangolo/uvicorn-gunicorn-docker/blob/master/docker-images/gunicorn_conf.py
 
-CMD ["uvicorn", "src.expasy_agent.api:app", "--host", "0.0.0.0", "--port", "80", "--workers", "8"]
+CMD ["uvicorn", "src.expasy_agent.api:app", "--host", "0.0.0.0", "--port", "80", "--workers", "6"]
 # CMD ["uvicorn", "src.expasy_agent.api", "--host", "0.0.0.0", "--port", "80", "--workers", "4", "--http", "h11"]

chat-with-context/demo/index.html

@@ -13,6 +13,7 @@
     <link href="./style.css" rel="stylesheet" />
 
     <script type="module" src="../src/chat-with-context.tsx"></script>
+    <link href="../src/style.css" rel="stylesheet" />
 
     <!-- <link href="https://unpkg.com/@sib-swiss/chat-with-context/dist/style.css" rel="stylesheet" />
     <script type="module" src="https://unpkg.com/@sib-swiss/chat-with-context/dist/chat-with-context.js"></script> -->
@@ -49,18 +50,20 @@ <h2 class="text-xl text-center font-semibold border-b">
           </div>
         </h2>
         <p>
-          <strong>ExpasyGPT</strong> is reshaping how researchers explore life science data, streamlining searches across
-          the <a href="https://www.sib.swiss/">SIB Swiss Institute of Bioinformatics</a>’ databases (learn more
+          <strong>ExpasyGPT</strong> is reshaping how researchers explore life science data, streamlining searches
+          across the <a href="https://www.sib.swiss/">SIB Swiss Institute of Bioinformatics</a>’ databases (learn more
           <a href="/about-chat">here</a>). This AI-driven tool is powered by metadata and example queries from SPARQL
           endpoints like:
         </p>
-        <ul style="list-style: inside;">
+        <ul style="list-style: inside">
           <li>
             <a href="http://www.uniprot.org/uniprot/" target="_blank">UniProt</a>, an expertly curated database of
             proteins
           </li>
           <li><a href="https://omabrowser.org/oma/home/" target="_blank">OMA</a>, the orthologous matrix</li>
-          <li><a href="https://www.bgee.org/" target="_blank">Bgee</a>, an expertly curated gene expression database</li>
+          <li>
+            <a href="https://www.bgee.org/" target="_blank">Bgee</a>, an expertly curated gene expression database
+          </li>
           <li>
             <a href="https://www.rhea-db.org/" target="_blank">Rhea</a>, an expertly curated database of biochemical
             reactions

chat-with-context/index.html

@@ -49,18 +49,20 @@ <h2 class="text-xl text-center font-semibold border-b pb-2">
           </div>
         </h2>
         <p>
-          <strong>ExpasyGPT</strong> is reshaping how researchers explore life science data, streamlining searches across
-          the <a href="https://www.sib.swiss/">SIB Swiss Institute of Bioinformatics</a>’ databases (learn more
+          <strong>ExpasyGPT</strong> is reshaping how researchers explore life science data, streamlining searches
+          across the <a href="https://www.sib.swiss/">SIB Swiss Institute of Bioinformatics</a>’ databases (learn more
           <a href="/about-chat">here</a>). This AI-driven tool is powered by metadata and example queries from SPARQL
           endpoints like:
         </p>
-        <ul style="list-style: inside;">
+        <ul style="list-style: inside">
           <li>
             <a href="http://www.uniprot.org/uniprot/" target="_blank">UniProt</a>, an expertly curated database of
             proteins
           </li>
           <li><a href="https://omabrowser.org/oma/home/" target="_blank">OMA</a>, the orthologous matrix</li>
-          <li><a href="https://www.bgee.org/" target="_blank">Bgee</a>, an expertly curated gene expression database</li>
+          <li>
+            <a href="https://www.bgee.org/" target="_blank">Bgee</a>, an expertly curated gene expression database
+          </li>
           <li>
             <a href="https://www.rhea-db.org/" target="_blank">Rhea</a>, an expertly curated database of biochemical
             reactions

chat-with-context/package.json

@@ -28,13 +28,10 @@
     "version": "npm run lint && npm run build && npm publish"
   },
   "dependencies": {
-    "@ai-sdk/openai": "^1.0.18",
     "@langchain/core": "^0.3.27",
     "@langchain/langgraph": "^0.2.39",
     "@langchain/langgraph-sdk": "^0.0.33",
-    "ai": "^4.0.33",
     "dompurify": "^3.2.3",
-    "feather-icons": "^4.29.2",
     "highlight.js": "^11.10.0",
     "marked": "^15.0.3",
     "solid-element": "^1.9.1",
@@ -47,7 +44,6 @@
     "@testing-library/dom": "^10.4.0",
     "@testing-library/jest-dom": "^6.6.3",
     "@types/eslint__js": "^8.42.3",
-    "@types/feather-icons": "^4.29.4",
     "@types/node": "^22.10.2",
     "@types/sparqljs": "^3.1.12",
     "auto-changelog": "^2.5.0",

chat-with-context/src/chat-with-context.tsx

@@ -20,7 +20,6 @@ import {streamResponse, ChatState} from "./providers";
 // SolidJS custom element: https://github.com/solidjs/solid/blob/main/packages/solid-element/README.md
 // https://github.com/solidjs/templates/tree/main/ts-tailwindcss
 
-
 /**
  * Custom element to create a chat interface with a context-aware assistant.
  * @example <chat-with-context api="http://localhost:8000/"></chat-with-context>
@@ -107,7 +106,9 @@ customElement("chat-with-context", {api: "", examples: "", apiKey: "", feedbackA
   }
 
   return (
-    <div class={`chat-with-context w-full h-full flex flex-col ${state.messages().length === 0 ? "justify-center" : ""}`}>
+    <div
+      class={`chat-with-context w-full h-full flex flex-col ${state.messages().length === 0 ? "justify-center" : ""}`}
+    >
       <style>{style}</style>
       {/* Main chat container */}
       <div ref={chatContainerEl} class={`overflow-y-auto ${state.messages().length !== 0 ? "flex-grow" : ""}`}>

chat-with-context/src/providers.ts

@@ -1,8 +1,5 @@
 import {Accessor, createSignal, Setter} from "solid-js";
 import {Client} from "@langchain/langgraph-sdk";
-import {RemoteRunnable} from "@langchain/core/runnables/remote";
-import {streamText} from 'ai';
-import {createOpenAI} from '@ai-sdk/openai';
 // import { RemoteGraph } from "@langchain/langgraph/remote";
 // import { isAIMessageChunk } from "@langchain/core/messages";
 
@@ -52,11 +49,7 @@ export class ChatState {
   setMessages: Setter<Message[]>;
   abortController: AbortController;
 
-  constructor({apiUrl, apiKey, model}: {
-    apiUrl: string;
-    apiKey: string;
-    model: string;
-  }) {
+  constructor({apiUrl, apiKey, model}: {apiUrl: string; apiKey: string; model: string}) {
     this.apiUrl = apiUrl.endsWith("/") ? apiUrl : apiUrl + "/";
     this.apiKey = apiKey;
     this.model = model;
@@ -70,7 +63,7 @@ export class ChatState {
   abortRequest = () => {
     this.abortController.abort();
     this.abortController = new AbortController();
-  }
+  };
 
   lastMsg = () => this.messages()[this.messages().length - 1];
 
@@ -102,23 +95,18 @@ export class ChatState {
   };
 }
 
-
-
 // Stream a response from various LLM agent providers (OpenAI-like, LangGraph, LangServe)
 export async function streamResponse(state: ChatState, question: string) {
   state.appendMessage(question, "user");
   if (state.apiUrl.endsWith(":2024/") || state.apiUrl.endsWith(":8123/")) {
     // Query LangGraph API
     await streamLangGraphApi(state);
-
-  } else if (state.apiUrl.endsWith("/langgraph/")) {
-    // Query LangGraph through LangServe API
-    await streamCustomLangGraph(state);
-
-  } else {
+  } else if (state.apiUrl.endsWith("/completions/")) {
     // Query the OpenAI-compatible chat API
     await streamOpenAILikeApi(state);
-    // await streamCustomLangGraph(statse);
+  } else {
+    // Query LangGraph through our custom API
+    await streamCustomLangGraph(state);
   }
 }
 
@@ -145,7 +133,10 @@ async function processLangGraphChunk(state: ChatState, chunk: any) {
         if (nodeData.extracted_entities.sparql_query) {
           state.lastMsg().setLinks([
             {
-              url: getEditorUrl(nodeData.extracted_entities.sparql_query, nodeData.extracted_entities.sparql_endpoint_url),
+              url: getEditorUrl(
+                nodeData.extracted_entities.sparql_query,
+                nodeData.extracted_entities.sparql_endpoint_url,
+              ),
               ...queryLinkLabels,
             },
           ]);
@@ -155,12 +146,7 @@ async function processLangGraphChunk(state: ChatState, chunk: any) {
       if (nodeData.validation) {
         for (const validationStep of nodeData.validation) {
           // Handle messages related to tools (includes post generation validation)
-          state.appendStepToLastMsg(
-            validationStep.label,
-            nodeId,
-            [],
-            validationStep.details,
-          );
+          state.appendStepToLastMsg(validationStep.label, nodeId, [], validationStep.details);
           if (validationStep.type === "recall") {
             // When recall-model is called, the model will re-generate the response, so we need to update the message
             // If the update contains a message with a fix (e.g. done during post generation validation)
@@ -185,7 +171,6 @@ async function processLangGraphChunk(state: ChatState, chunk: any) {
     if (msg.content && msg.type === "tool") {
       // If tool called by model
       state.appendStepToLastMsg(`🔧 Tool ${msg.name} result: ${msg.content}`, metadata.langgraph_node);
-
     } else if (msg.content && msg.type === "AIMessageChunk") {
       // Response from the model
       state.appendContentToLastMsg(msg.content);
@@ -198,7 +183,7 @@ async function streamCustomLangGraph(state: ChatState) {
     method: "POST",
     headers: {
       "Content-Type": "application/json",
-      Authorization: `Bearer ${state.apiKey}`
+      Authorization: `Bearer ${state.apiKey}`,
     },
     signal: state.abortController.signal,
     body: JSON.stringify({
@@ -229,7 +214,7 @@ async function streamCustomLangGraph(state: ChatState) {
         // console.log(line)
         try {
           const json = JSON.parse(line);
-          processLangGraphChunk(state, json)
+          processLangGraphChunk(state, json);
           partialLine = "";
         } catch {
           partialLine += line;
@@ -261,13 +246,12 @@ async function streamLangGraphApi(state: ChatState) {
   }
 }
 
-
 async function streamOpenAILikeApi(state: ChatState) {
   const response = await fetch(`${state.apiUrl}chat/completions`, {
     method: "POST",
     headers: {
       "Content-Type": "application/json",
-      Authorization: `Bearer ${state.apiKey}`
+      Authorization: `Bearer ${state.apiKey}`,
     },
     signal: state.abortController.signal,
     body: JSON.stringify({

chat-with-context/src/utils.tsx

@@ -1,4 +1,7 @@
-export const queryLinkLabels = {label: "Run and edit the query", title: "Open the SPARQL query in an editor in a new tab"};
+export const queryLinkLabels = {
+  label: "Run and edit the query",
+  title: "Open the SPARQL query in an editor in a new tab",
+};
 
 export function extractSparqlQuery(markdownContent: string) {
   // Regular expression to match SPARQL queries within code blocks
@@ -13,7 +16,7 @@ export function extractSparqlQuery(markdownContent: string) {
   const endpointMatch = lastQuery.match(endpointRegex);
   const endpointUrl = endpointMatch ? endpointMatch[1] : null;
   if (!endpointUrl) return null;
-  return getEditorUrl(lastQuery, endpointUrl)
+  return getEditorUrl(lastQuery, endpointUrl);
   // return <a
   //   class="my-3 px-3 py-1 text-sm text-black dark:text-white bg-gray-300 hover:bg-gray-400 dark:bg-gray-700 dark:hover:bg-gray-800 rounded-lg"
   //   href="https://sib-swiss.github.io/sparql-editor/?endpoint=${endpointUrl}&query=${encodeURIComponent(lastQuery)}"
@@ -24,7 +27,7 @@ export function extractSparqlQuery(markdownContent: string) {
 }
 
 export function getEditorUrl(query: string, endpointUrl: string = "") {
-  return `https://sib-swiss.github.io/sparql-editor/?${(endpointUrl) ? `endpoint=${endpointUrl}&` : ""}query=${encodeURIComponent(query)}`;
+  return `https://sib-swiss.github.io/sparql-editor/?${endpointUrl ? `endpoint=${endpointUrl}&` : ""}query=${encodeURIComponent(query)}`;
 }
 
 export function getLangForDocType(docType: string) {
@@ -50,9 +53,10 @@ export const style = `chat-with-context {
 }
 .iconBtn {
   filter: invert(44%) sepia(22%) saturate(496%) hue-rotate(176deg) brightness(93%) contrast(79%);
-}
-@keyframes spin {
-  to {
-    transform: rotate(360deg);
-  }
 }`;
+
+// @keyframes spin {
+//   to {
+//     transform: rotate(360deg);
+//   }
+// }

compose.dev.yml

@@ -16,11 +16,9 @@ services:
       - 8000:80
     volumes:
       - ./packages:/app/packages
+    environment:
+      - VECTORDB_HOST=vectordb
     command: ["uvicorn", "src.expasy_agent.api:app", "--host", "0.0.0.0", "--port", "80", "--reload"]
-    # command: ["uvicorn", "src.expasy_agent.api:app", "--host", "0.0.0.0", "--port", "80", "--http", "httptools"]
-    # command: ["uvicorn", "src.expasy_agent.api:app", "--host", "0.0.0.0", "--port", "80", "--http", "h11"]
-    # h11 written in python, httptools in C
-    # entrypoint: /start-reload.sh --http h11
 
   # langgraph-redis:
   #   extends:

compose.yml

@@ -10,9 +10,6 @@ services:
       # - ./qdrant_config.yml:/qdrant/config/production.yaml
     environment:
       - QDRANT_ALLOW_RECOVERY_MODE=true
-    ports:
-      - 6333:6333
-      - 6334:6334
     # command:
     #   - ./qdrant --config-path /qdrant/config/production.yaml
 
@@ -25,8 +22,9 @@ services:
     ports:
       - 443:80
     environment:
-      - MAX_WORKERS=6
-      # cf. https://github.com/tiangolo/uvicorn-gunicorn-docker/blob/master/docker-images/gunicorn_conf.py
+      # NOTE: disguting hack to fix a bug with podman internal network (podman devs are not competent, this bug comes and goes)
+      - VECTORDB_HOST=10.89.1.2
+      # https://github.com/tiangolo/uvicorn-gunicorn-docker/blob/master/docker-images/gunicorn_conf.py
     volumes:
       - ./data/fastembed_cache:/tmp/fastembed_cache
       - ./data/logs:/logs

packages/expasy-agent/src/expasy_agent/api.py

@@ -120,7 +120,7 @@ def stream_dict(d: dict) -> str:
 
 
 # @app.post("/chat/completions")
-@app.post("/langgraph")
+@app.post("/chat")
 async def chat(request: Request):
     """Chat with the assistant main endpoint."""
     auth_header = request.headers.get("Authorization")
@@ -263,13 +263,13 @@ def get_user_logs(request: LogsRequest):
 
 @app.get("/", response_class=HTMLResponse, include_in_schema=False)
 def chat_ui(request: Request) -> Any:
-    """Render the chat UI using jinja2 + HTML"""
+    """Render the chat UI using jinja2 + HTML."""
     return templates.TemplateResponse(
         "index.html",
         {
             "request": request,
             "expasy_key": settings.expasy_api_key,
-            "api_url": "https://chat.expasy.org/langgraph/",
+            "api_url": "https://chat.expasy.org/chat/",
             "examples": ",".join([
                 "Which resources are available at the SIB?",
                 "How can I get the HGNC symbol for the protein P68871?",

packages/expasy-agent/src/expasy_agent/config.py

@@ -10,7 +10,6 @@
 from openai import OpenAI
 from pydantic_settings import BaseSettings, SettingsConfigDict
 from qdrant_client import QdrantClient
-from sparql_llm.utils import get_prefixes_for_endpoints
 
 from expasy_agent import prompts
 
@@ -37,10 +36,10 @@ class Settings(BaseSettings):
     ontology_chunk_size: int = 3000
     ontology_chunk_overlap: int = 200
 
-    # vectordb_host: str = "vectordb"
+    vectordb_host: str = "vectordb"
     # vectordb_host: str = "localhost"
-    # NOTE: old hack to fix a bug with podman internal network, can be removed soon
-    vectordb_host: str = "10.89.1.2"
+    # NOTE: disguting hack to fix a bug with podman internal network
+    # vectordb_host: str = "10.89.1.2"
 
     docs_collection_name: str = "expasy"
     entities_collection_name: str = "entities"

packages/expasy-agent/src/expasy_agent/indexing/index_wikidata.py

@@ -0,0 +1 @@
+"""Index Wikidata endpoint to be used for QALD-10 benchmark."""