From 969d25209ca52f4a0866ccc2362f6a8902a84fc2 Mon Sep 17 00:00:00 2001
From: idchlife <idchlife@gmail.com>
Date: Mon, 9 Mar 2026 09:59:27 +0300
Subject: [PATCH] using openailike for llamaindex that will allow of usage of
 any openai compatible models via api, and not just hardcoded models inside
 default OpenAI class in llamaindex

---
 services/rag/.DS_Store                        | Bin 8196 -> 8196 bytes
 services/rag/llamaindex/.env.dist             |   1 +
 services/rag/llamaindex/PLANNING.md           |   7 ++++
 services/rag/llamaindex/config.py             |  18 +++++++--
 .../helpers/openai_compatible_llm.py          |  35 ++++++++++++++++++
 services/rag/llamaindex/requirements.txt      |   1 +
 6 files changed, 59 insertions(+), 3 deletions(-)
 create mode 100644 services/rag/llamaindex/helpers/openai_compatible_llm.py

diff --git a/services/rag/.DS_Store b/services/rag/.DS_Store
index 5bd78c607c9e3da706e2985d1f0f69fe1639eccb..5c7d7c4a5b90ff7c9b645452de3f3b520324e874 100644
GIT binary patch
delta 267
zcmZp1XmOa}FUrEez`)4BAi%(o%#gv5$dJj9msGy7a2or>2Hwr=94s7+ER!3B#p;vF
ziwlx+@{@qdcO(_$WEPhg7+hmyVrF4wW9MM!;O2-8&d4thE=epYEp|#QiU#pQ^7C_&
zVC=-Cu*}r*cmWaT{JfIH%)Hbhu!hW(RG^ZW@XWlF{Bo!K(!7*nu-;&Z1P3PvXS{$!
zwQIGRxrL5`p|N?bjzYDik%5kaiLqI2EhmSlvc7dte0EN5UVazQaX`Qb^bZ*DLTMP)
c4fKiRW>o<W_RZ`P-&iKE6L8y{DRPhr0EyE_=>Px#

delta 67
zcmZp1XmOa}FUrioz`)4BAi%(o%#gv5$dJj9ms7m4a2or>2Hwr=94s7+%#))<Bsc#T
W-~bBU+RQHTjb-z7Q3Ixl4c7rD{1biv

diff --git a/services/rag/llamaindex/.env.dist b/services/rag/llamaindex/.env.dist
index 2d5a3bb..d6a3dd5 100644
--- a/services/rag/llamaindex/.env.dist
+++ b/services/rag/llamaindex/.env.dist
@@ -14,6 +14,7 @@ QDRANT_GRPC_PORT=6334
 # OpenAI Configuration (for reference - uncomment and configure when using OpenAI strategy)
 # OPENAI_CHAT_URL=https://api.openai.com/v1
 # OPENAI_CHAT_KEY=your_openai_api_key_here
+# OPENAI_CHAT_IS_FUNCTION_CALLING_MODEL=false
 # OPENAI_EMBEDDING_MODEL=text-embedding-3-small
 # OPENAI_EMBEDDING_BASE_URL=https://api.openai.com/v1
 # OPENAI_EMBEDDING_API_KEY=your_openai_api_key_here
diff --git a/services/rag/llamaindex/PLANNING.md b/services/rag/llamaindex/PLANNING.md
index 65cad9a..e84cfcd 100644
--- a/services/rag/llamaindex/PLANNING.md
+++ b/services/rag/llamaindex/PLANNING.md
@@ -93,3 +93,10 @@ Chosen data folder: relatve ./../../../data - from the current folder
   - optionally filter low-information chunks (headers/footers)
 - [x] Add optional metadata-aware retrieval improvements (years/events/keywords) parity with LangChain approach (folder near current folder), if feasible in the chosen LlamaIndex primitives.
 - [x] Update `server.py` endpoint to use the new agent-like chat path (keep simple retrieval endpoint available as fallback or debug mode).
+
+# Phase 13 (make wrapper around OpenAI llamaindex class or install library to help use any openai compatible models)
+
+Inside config.py file object is being created for using as Chat Model. Unfortunately, it allows only "supported" models as value for "model" argument in constructor.
+
+- [x] Search online for library or plugin for llamainde, that fixes the OpenAI class behaviour or provides replacement, that will allow any models. If found, install it in project venv, update then requirements.txt and replace usage of OpenAI to new one in the code.
+- [x] Fallback option: create wrapper of OpenAI class, that will inherit and replace methods/features that check for "registered" models. Use the replacement then in the code.
diff --git a/services/rag/llamaindex/config.py b/services/rag/llamaindex/config.py
index 9348aa7..e948e90 100644
--- a/services/rag/llamaindex/config.py
+++ b/services/rag/llamaindex/config.py
@@ -93,18 +93,30 @@ def get_llm_model():
         return llm
 
     elif strategy == "openai":
-        from llama_index.llms.openai import OpenAI
+        from llama_index.llms.openai_like import OpenAILike
+        # from helpers.openai_compatible_llm import OpenAICompatibleLLM
 
         openai_chat_url = os.getenv("OPENAI_CHAT_URL", "https://api.openai.com/v1")
         openai_chat_key = os.getenv("OPENAI_CHAT_KEY", "dummy_key_for_template")
         openai_chat_model = os.getenv("OPENAI_CHAT_MODEL", "gpt-3.5-turbo")
+        openai_is_fc_model = (
+            os.getenv("OPENAI_CHAT_IS_FUNCTION_CALLING_MODEL", "false").lower()
+            == "true"
+        )
 
         # Set the API key in environment for OpenAI
         os.environ["OPENAI_API_KEY"] = openai_chat_key
 
-        logger.info(f"Initializing OpenAI chat model: {openai_chat_model}")
+        logger.info(
+            f"Initializing OpenAI-compatible chat model: {openai_chat_model} "
+            f"(base={openai_chat_url}, function_calling={openai_is_fc_model})"
+        )
 
-        llm = OpenAI(model=openai_chat_model, api_base=openai_chat_url)
+        llm = OpenAILike(
+            model=openai_chat_model,
+            api_base=openai_chat_url,
+            api_key=openai_chat_key,
+        )
 
         return llm
 
diff --git a/services/rag/llamaindex/helpers/openai_compatible_llm.py b/services/rag/llamaindex/helpers/openai_compatible_llm.py
new file mode 100644
index 0000000..a8778cf
--- /dev/null
+++ b/services/rag/llamaindex/helpers/openai_compatible_llm.py
@@ -0,0 +1,35 @@
+"""
+OpenAI-compatible LLM wrapper for LlamaIndex chat models.
+
+This wrapper is used as a fallback/replacement for strict OpenAI model validation paths.
+It relies on LlamaIndex `OpenAILike`, which supports arbitrary model names for
+OpenAI-compatible endpoints.
+"""
+
+from llama_index.llms.openai_like import OpenAILike
+
+
+class OpenAICompatibleLLM(OpenAILike):
+    """
+    Thin wrapper over OpenAILike with chat-friendly defaults.
+    """
+
+    def __init__(
+        self,
+        model: str,
+        api_base: str,
+        api_key: str,
+        temperature: float = 0.1,
+        timeout: float = 120.0,
+        is_function_calling_model: bool = False,
+    ):
+        super().__init__(
+            model=model,
+            api_base=api_base,
+            api_key=api_key,
+            temperature=temperature,
+            timeout=timeout,
+            # Explicitly avoid "registered model only" assumptions.
+            is_chat_model=True,
+            is_function_calling_model=is_function_calling_model,
+        )
diff --git a/services/rag/llamaindex/requirements.txt b/services/rag/llamaindex/requirements.txt
index 3196401..9b08e74 100644
--- a/services/rag/llamaindex/requirements.txt
+++ b/services/rag/llamaindex/requirements.txt
@@ -74,6 +74,7 @@ llama-index-indices-managed-llama-cloud==0.9.4
 llama-index-instrumentation==0.4.2
 llama-index-llms-ollama==0.9.1
 llama-index-llms-openai==0.6.17
+llama-index-llms-openai-like==0.6.0
 llama-index-readers-file==0.5.6
 llama-index-readers-llama-parse==0.5.1
 llama-index-vector-stores-qdrant==0.9.1