Merge pull request #28 from IABTechLab/feature/integrate-llm-providers

Feature/integrate llm providers

Author: therevoltingx

.env.example

@@ -147,11 +147,61 @@ PREFERRED_DEAL_DISCOUNT_MAX=0.15
 # =============================================================================
 # LLM Configuration
 # =============================================================================
+# Only one provider can be active at a time.
+# Uncomment the block for your chosen provider and comment out the others.
+
+# Provider: Anthropic (default)
+ANTHROPIC_API_KEY=your-anthropic-api-key-here
 DEFAULT_LLM_MODEL=anthropic/claude-sonnet-4-5-20250929
-MANAGER_LLM_MODEL=anthropic/claude-sonnet-4-5-20250929
+MANAGER_LLM_MODEL=anthropic/claude-opus-4-20250514
 LLM_TEMPERATURE=0.3
 LLM_MAX_TOKENS=4096
 
+# Provider: OpenAI (alternative)
+# To switch to OpenAI, set DEFAULT_LLM_MODEL and OPENAI_API_KEY:
+# OPENAI_API_KEY=your-openai-api-key-here
+# DEFAULT_LLM_MODEL=openai/gpt-4o
+# MANAGER_LLM_MODEL=openai/gpt-4o
+
+# Provider: Google Gemini (alternative)
+# To switch to Gemini, set DEFAULT_LLM_MODEL and GOOGLE_API_KEY:
+# GOOGLE_API_KEY=your-google-api-key-here
+# DEFAULT_LLM_MODEL=gemini/gemini-2.5-flash
+# MANAGER_LLM_MODEL=gemini/gemini-2.5-flash
+
+# Provider: AWS Bedrock (alternative)
+# To switch to Bedrock, set DEFAULT_LLM_MODEL and AWS credentials:
+# AWS_ACCESS_KEY_ID=your-aws-access-key
+# AWS_SECRET_ACCESS_KEY=your-aws-secret-key
+# AWS_REGION=us-west-2
+# DEFAULT_LLM_MODEL=bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0
+# MANAGER_LLM_MODEL=bedrock/us.anthropic.claude-opus-4-20250514-v1:0
+
+# Provider: Custom OpenAI-compatible endpoint (alternative)
+# For endpoints with no CrewAI native provider prefix of their own — NVIDIA
+# NIM, Ollama, HuggingFace TGI, vLLM, LM Studio, etc. Set DEFAULT_LLM_MODEL to
+# the raw model id the endpoint expects, plus
+# OPENAI_COMPATIBLE_LLM_API_BASE_URL. OPENAI_COMPATIBLE_LLM_API_KEY is
+# optional — omit it for endpoints (e.g. a local Ollama server) that don't
+# require one.
+#
+# NVIDIA NIM:
+# OPENAI_COMPATIBLE_LLM_API_KEY=nvapi-your-nvidia-api-key
+# OPENAI_COMPATIBLE_LLM_API_BASE_URL=https://integrate.api.nvidia.com/v1
+# DEFAULT_LLM_MODEL=meta/llama-3.1-70b-instruct
+# MANAGER_LLM_MODEL=meta/llama-3.1-70b-instruct
+#
+# Ollama (local or self-hosted):
+# OPENAI_COMPATIBLE_LLM_API_BASE_URL=http://localhost:11434/v1
+# DEFAULT_LLM_MODEL=llama3
+# MANAGER_LLM_MODEL=llama3
+#
+# HuggingFace TGI (self-hosted):
+# OPENAI_COMPATIBLE_LLM_API_KEY=hf_your-huggingface-token
+# OPENAI_COMPATIBLE_LLM_API_BASE_URL=https://your-tgi-endpoint.example.com/v1
+# DEFAULT_LLM_MODEL=meta-llama/Meta-Llama-3-8B-Instruct
+# MANAGER_LLM_MODEL=meta-llama/Meta-Llama-3-8B-Instruct
+
 # =============================================================================
 # CrewAI Configuration
 # =============================================================================

docs/guides/configuration.md

@@ -89,6 +89,8 @@ ad-seller freewheel-login --provider bc
 | `MANAGER_LLM_MODEL` | `str` | `"anthropic/claude-opus-4-20250514"` | Model for manager/orchestrator agents |
 | `LLM_TEMPERATURE` | `float` | `0.3` | LLM temperature (lower = more deterministic) |
 | `LLM_MAX_TOKENS` | `int` | `4096` | Maximum tokens per LLM response |
+| `OPENAI_COMPATIBLE_LLM_API_KEY` | `str` | `None` | API key for a [custom OpenAI-compatible endpoint](#custom-openai-compatible-endpoints) (optional — some endpoints don't require one) |
+| `OPENAI_COMPATIBLE_LLM_API_BASE_URL` | `str` | `None` | Base URL for a custom OpenAI-compatible endpoint; when set, `DEFAULT_LLM_MODEL`/`MANAGER_LLM_MODEL` are sent as-is to that endpoint instead of being routed by provider prefix |
 
 ### Supported Providers
 
@@ -98,7 +100,7 @@ The seller agent uses CrewAI's native provider integrations via `provider/model-
 |----------|-------------|-----------------|---------------|
 | **Anthropic** (default) | `anthropic/claude-sonnet-4-5-20250929` | `ANTHROPIC_API_KEY` | `crewai[anthropic]` |
 | **OpenAI** | `openai/gpt-4o` | `OPENAI_API_KEY` | `crewai[openai]` |
-| **Google Gemini** | `gemini/gemini-2.0-flash` | `GOOGLE_API_KEY` | `crewai[gemini]` |
+| **Google Gemini** | `gemini/gemini-2.5-flash` | `GOOGLE_API_KEY` | `crewai[gemini]` |
 | **Azure OpenAI** | `azure/my-deployment` | `AZURE_API_KEY`, `AZURE_API_BASE` | `crewai[azure]` |
 | **AWS Bedrock** | `bedrock/anthropic.claude-3-sonnet` | AWS credentials | `crewai[bedrock]` |
 
@@ -116,6 +118,33 @@ OPENAI_API_KEY=sk-xxxxx
 
 For other providers, see the [CrewAI LLM documentation](https://docs.crewai.com/en/learn/litellm-removal-guide).
 
+### Custom OpenAI-Compatible Endpoints
+
+For endpoints that don't have one of the native provider prefixes above —
+NVIDIA NIM, Ollama, HuggingFace TGI, vLLM, LM Studio, and similar — set
+`OPENAI_COMPATIBLE_LLM_API_BASE_URL`. This pins the request to CrewAI's native OpenAI-compatible
+client regardless of the model id's shape, so `DEFAULT_LLM_MODEL` /
+`MANAGER_LLM_MODEL` should be set to the raw model id the endpoint expects
+(no provider prefix). `OPENAI_COMPATIBLE_LLM_API_KEY` is optional — omit it for endpoints, such
+as a local Ollama server, that don't require one.
+
+**Example — NVIDIA NIM:**
+
+```bash
+OPENAI_COMPATIBLE_LLM_API_KEY=nvapi-xxxxx
+OPENAI_COMPATIBLE_LLM_API_BASE_URL=https://integrate.api.nvidia.com/v1
+DEFAULT_LLM_MODEL=meta/llama-3.1-70b-instruct
+MANAGER_LLM_MODEL=meta/llama-3.1-70b-instruct
+```
+
+**Example — Ollama, local or self-hosted:**
+
+```bash
+OPENAI_COMPATIBLE_LLM_API_BASE_URL=http://localhost:11434/v1
+DEFAULT_LLM_MODEL=llama3
+MANAGER_LLM_MODEL=llama3
+```
+
 ## Database & Storage
 
 | Variable | Type | Default | Description |

pyproject.toml

@@ -19,7 +19,7 @@ classifiers = [
 ]
 
 dependencies = [
-    "crewai[anthropic,bedrock]>=1.14.4,<2.0.0",
+    "crewai[anthropic,bedrock,openai,gemini]>=1.14.4,<2.0.0",
     "crewai-tools>=1.14.0,<2.0.0",
     "pydantic>=2.0.0",
     "pydantic-settings>=2.0.0",

src/ad_seller/agents/level1/inventory_manager.py

@@ -8,9 +8,10 @@
 function, evaluates proposals, and coordinates specialist agents.
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_inventory_manager() -> Agent:
@@ -28,7 +29,7 @@ def create_inventory_manager() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.manager_llm_model,
         temperature=0.3,
         max_tokens=settings.llm_max_tokens,

src/ad_seller/agents/level2/ctv_inventory_agent.py

@@ -7,9 +7,10 @@
 OTT apps, FAST channels, and household-targeted inventory.
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_ctv_inventory_agent() -> Agent:
@@ -27,7 +28,7 @@ def create_ctv_inventory_agent() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.default_llm_model,
         temperature=0.5,
         max_tokens=settings.llm_max_tokens,

src/ad_seller/agents/level2/display_inventory_agent.py

@@ -7,9 +7,10 @@
 and premium homepage takeovers.
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_display_inventory_agent() -> Agent:
@@ -26,7 +27,7 @@ def create_display_inventory_agent() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.default_llm_model,
         temperature=0.5,
         max_tokens=settings.llm_max_tokens,

src/ad_seller/agents/level2/linear_tv_inventory_agent.py

@@ -19,9 +19,10 @@
   integration lands)
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_linear_tv_inventory_agent() -> Agent:
@@ -41,7 +42,7 @@ def create_linear_tv_inventory_agent() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.default_llm_model,
         temperature=0.5,
         max_tokens=settings.llm_max_tokens,

src/ad_seller/agents/level2/mobile_app_inventory_agent.py

@@ -7,9 +7,10 @@
 SDK inventory, rewarded video, interstitials, and in-app bidding.
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_mobile_app_inventory_agent() -> Agent:
@@ -29,7 +30,7 @@ def create_mobile_app_inventory_agent() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.default_llm_model,
         temperature=0.5,
         max_tokens=settings.llm_max_tokens,

src/ad_seller/agents/level2/native_inventory_agent.py

@@ -7,9 +7,10 @@
 content recommendations, and sponsored content.
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_native_inventory_agent() -> Agent:
@@ -27,7 +28,7 @@ def create_native_inventory_agent() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.default_llm_model,
         temperature=0.5,
         max_tokens=settings.llm_max_tokens,

src/ad_seller/agents/level2/video_inventory_agent.py

@@ -7,9 +7,10 @@
 and outstream formats.
 """
 
-from crewai import LLM, Agent
+from crewai import Agent
 
 from ...config import get_settings
+from ...llm import build_llm
 
 
 def create_video_inventory_agent() -> Agent:
@@ -26,7 +27,7 @@ def create_video_inventory_agent() -> Agent:
     """
     settings = get_settings()
 
-    llm = LLM(
+    llm = build_llm(
         model=settings.default_llm_model,
         temperature=0.5,
         max_tokens=settings.llm_max_tokens,