intelligence-suite 0.2.2__tar.gz → 0.2.4__tar.gz

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/CodeIntelligence/rag_server.py

@@ -8,12 +8,18 @@ from intelligence_core.embedder import get_embedder
 from intelligence_core.store import ChromaStore
 from intelligence_core.server_base import create_app
 from intelligence_core.config import settings
+from intelligence_core.llm import get_module_llm_provider
 
 
 def build_app():
     store     = ChromaStore(collection_name="code_intelligence")
     retriever = Retriever(embedder=get_embedder(), store=store)
-    return create_app(title="CodeIntelligence RAG Server", retriever=retriever, module="code")
+    return create_app(
+        title="CodeIntelligence RAG Server",
+        retriever=retriever,
+        module="code",
+        llm_provider=get_module_llm_provider("ci"),
+    )
 
 
 app = build_app()

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/DocIntelligence/doc_server.py

@@ -8,12 +8,18 @@ from intelligence_core.store import ChromaStore
 from intelligence_core.embedder import get_embedder
 from intelligence_core.server_base import create_app
 from intelligence_core.config import settings
+from intelligence_core.llm import get_module_llm_provider
 
 
 def build_app():
     store = ChromaStore(collection_name="doc_intelligence")
     retriever = Retriever(embedder=get_embedder(), store=store)
-    return create_app(title="DocIntelligence RAG Server", retriever=retriever, module="doc")
+    return create_app(
+        title="DocIntelligence RAG Server",
+        retriever=retriever,
+        module="doc",
+        llm_provider=get_module_llm_provider("di"),
+    )
 
 
 app = build_app()

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/MentorIntelligence/mentor_server.py

@@ -86,7 +86,13 @@ def build_app() -> FastAPI:
     orchestrator = MentorOrchestrator(code_retriever, doc_retriever, mentor_retriever)
 
     base_retriever = Retriever(embedder=embedder, store=ChromaStore("mentor_intelligence"))
-    app = create_app(title="MentorIntelligence Server", retriever=base_retriever, module="mentor")
+    from intelligence_core.llm import get_module_llm_provider
+    app = create_app(
+        title="MentorIntelligence Server",
+        retriever=base_retriever,
+        module="mentor",
+        llm_provider=get_module_llm_provider("mi"),
+    )
 
     @app.post("/api/v1/mentor/onboard", response_model=OnboardResponse)
     def onboard(req: OnboardRequest):

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: intelligence-suite
-Version: 0.2.2
+Version: 0.2.4
 Summary: Modular knowledge retrieval suite: code, docs, APIs — all on-premise
 Author-email: ViciusLio <viciuslios@gmail.com>
 License-Expression: MIT
@@ -654,6 +654,28 @@ Switch backend with a single env var — no code changes required.
 > Any OpenAI-compatible server works with `LLM_BACKEND=openai` or `vllm` by pointing
 > `OPENAI_BASE_URL` at the correct endpoint.
 
+### Per-module LLM routing
+
+Each module can use a **different LLM backend and model** independently.
+Set any combination of `CI_LLM_*`, `DI_LLM_*`, `MI_LLM_*` in `.env` — leave empty to fall back to the global `LLM_BACKEND`:
+
+```env
+# CodeIntelligence → vLLM GPU server with code-specialised model
+CI_LLM_BACKEND=openai
+CI_LLM_MODEL=codellama:34b
+CI_LLM_BASE_URL=http://gpu-server:8000/v1
+
+# DocIntelligence → local Mistral (better multilingual / Italian)
+DI_LLM_BACKEND=ollama
+DI_LLM_MODEL=mistral:7b
+
+# MentorIntelligence → Claude API (best pedagogical quality)
+MI_LLM_BACKEND=claude
+MI_LLM_MODEL=claude-sonnet-4-5
+```
+
+Any OpenAI-compatible endpoint (vLLM, Groq, Mistral AI, LM Studio, Azure…) works by setting `*_LLM_BACKEND=openai` and `*_LLM_BASE_URL` to the endpoint.
+
 ### Escalation
 
 When retrieval confidence < `ESCALATION_THRESHOLD` and `ANTHROPIC_API_KEY` is set,

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/README.md

@@ -595,6 +595,28 @@ Switch backend with a single env var — no code changes required.
 > Any OpenAI-compatible server works with `LLM_BACKEND=openai` or `vllm` by pointing
 > `OPENAI_BASE_URL` at the correct endpoint.
 
+### Per-module LLM routing
+
+Each module can use a **different LLM backend and model** independently.
+Set any combination of `CI_LLM_*`, `DI_LLM_*`, `MI_LLM_*` in `.env` — leave empty to fall back to the global `LLM_BACKEND`:
+
+```env
+# CodeIntelligence → vLLM GPU server with code-specialised model
+CI_LLM_BACKEND=openai
+CI_LLM_MODEL=codellama:34b
+CI_LLM_BASE_URL=http://gpu-server:8000/v1
+
+# DocIntelligence → local Mistral (better multilingual / Italian)
+DI_LLM_BACKEND=ollama
+DI_LLM_MODEL=mistral:7b
+
+# MentorIntelligence → Claude API (best pedagogical quality)
+MI_LLM_BACKEND=claude
+MI_LLM_MODEL=claude-sonnet-4-5
+```
+
+Any OpenAI-compatible endpoint (vLLM, Groq, Mistral AI, LM Studio, Azure…) works by setting `*_LLM_BACKEND=openai` and `*_LLM_BASE_URL` to the endpoint.
+
 ### Escalation
 
 When retrieval confidence < `ESCALATION_THRESHOLD` and `ANTHROPIC_API_KEY` is set,

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/intelligence_core/config.py

@@ -56,6 +56,26 @@ class Settings(BaseSettings):
         """Expand ~ and resolve to an absolute path so CWD never matters."""
         return str(Path(v).expanduser().resolve())
 
+    # ── Per-module LLM overrides ───────────────────────────────────────────────
+    # Leave empty to use the global LLM_BACKEND / OLLAMA_MODEL / OPENAI_* settings.
+    # Set any combination to route a specific module to a different backend or model.
+    #
+    # CodeIntelligence  (CI_LLM_*)
+    ci_llm_backend:  str = ""   # e.g. "ollama" | "openai" | "vllm" | "claude"
+    ci_llm_model:    str = ""   # e.g. "qwen2.5-coder:7b"
+    ci_llm_base_url: str = ""   # e.g. "http://gpu-server:8000/v1"
+    ci_llm_api_key:  str = ""
+    # DocIntelligence   (DI_LLM_*)
+    di_llm_backend:  str = ""   # e.g. "ollama" | "openai" | "claude"
+    di_llm_model:    str = ""   # e.g. "mistral:7b"
+    di_llm_base_url: str = ""
+    di_llm_api_key:  str = ""
+    # MentorIntelligence (MI_LLM_*)
+    mi_llm_backend:  str = ""   # e.g. "claude"
+    mi_llm_model:    str = ""   # e.g. "claude-sonnet-4-5"
+    mi_llm_base_url: str = ""
+    mi_llm_api_key:  str = ""
+
     # ── Server ports (one per module, avoids conflicts when running together) ──
     ci_port: int = 8080   # CodeIntelligence
     di_port: int = 8081   # DocIntelligence

intelligence_suite-0.2.4/intelligence_core/llm/__init__.py

@@ -0,0 +1,145 @@
+"""LLM provider factory for Intelligence Suite.
+
+Supported backends
+------------------
+ollama       Local Ollama server (default, zero cost, no API key)
+openai       OpenAI Chat Completions API
+vllm         vLLM local GPU server (OpenAI-compatible)
+claude       Anthropic Claude API
+
+Any other OpenAI-compatible server (Groq, Mistral AI, LM Studio, Together AI, …)
+can be used with ``LLM_BACKEND=openai`` by changing ``OPENAI_BASE_URL``.
+
+Per-module routing
+------------------
+Each module can override the global backend/model independently:
+
+    CI_LLM_BACKEND=openai          CI_LLM_MODEL=codellama:34b
+    CI_LLM_BASE_URL=http://gpu:8000/v1
+
+    DI_LLM_BACKEND=ollama          DI_LLM_MODEL=mistral:7b
+
+    MI_LLM_BACKEND=claude          MI_LLM_MODEL=claude-sonnet-4-5
+
+Leave any variable empty to fall back to the global LLM_BACKEND settings.
+
+Usage
+-----
+    from intelligence_core.llm import get_llm_provider, get_module_llm_provider
+
+    llm = get_llm_provider()              # global settings
+    llm = get_module_llm_provider("ci")   # CodeIntelligence — with per-module override
+    llm = get_module_llm_provider("di")   # DocIntelligence
+    llm = get_module_llm_provider("mi")   # MentorIntelligence
+"""
+
+from __future__ import annotations
+import logging
+
+from intelligence_core.llm.protocol import LLMProvider, SYSTEM_PROMPT_DEFAULT
+
+logger = logging.getLogger(__name__)
+
+
+def get_llm_provider(
+    backend: str | None = None,
+    *,
+    model: str | None = None,
+    base_url: str | None = None,
+    api_key: str | None = None,
+) -> LLMProvider:
+    """
+    Factory: return the configured LLM provider.
+
+    Args:
+        backend:  Override ``LLM_BACKEND`` from settings.
+                  Values: ``"ollama"`` | ``"openai"`` | ``"vllm"`` | ``"claude"``
+        model:    Override the model name (OLLAMA_MODEL / OPENAI_MODEL / CLAUDE_MODEL).
+        base_url: Override the API base URL (OPENAI_BASE_URL / OLLAMA_BASE_URL).
+        api_key:  Override the API key (OPENAI_API_KEY / ANTHROPIC_API_KEY).
+
+    Returns:
+        An object satisfying the :class:`LLMProvider` protocol.
+    """
+    from intelligence_core.config import settings
+
+    _backend = backend or settings.llm_backend
+
+    if _backend in ("openai", "vllm"):
+        from intelligence_core.llm.openai_compat import OpenAICompatProvider
+        return OpenAICompatProvider(
+            base_url=base_url or settings.openai_base_url,
+            api_key=api_key or settings.openai_api_key,
+            model=model or settings.openai_model,
+            backend_hint=_backend,
+        )
+
+    if _backend == "claude":
+        from intelligence_core.llm.claude import ClaudeProvider
+        return ClaudeProvider(
+            api_key=api_key or settings.anthropic_api_key,
+            model=model or settings.claude_model,
+        )
+
+    # default: ollama
+    from intelligence_core.llm.ollama import OllamaProvider
+    return OllamaProvider(
+        base_url=base_url or settings.ollama_base_url,
+        model=model or settings.ollama_model,
+    )
+
+
+def get_module_llm_provider(module: str) -> LLMProvider:
+    """
+    Get the LLM provider for a specific module, applying per-module env overrides.
+
+    Module prefixes
+    ---------------
+    "ci"  → CodeIntelligence   (CI_LLM_BACKEND, CI_LLM_MODEL, CI_LLM_BASE_URL, CI_LLM_API_KEY)
+    "di"  → DocIntelligence    (DI_LLM_BACKEND, DI_LLM_MODEL, DI_LLM_BASE_URL, DI_LLM_API_KEY)
+    "mi"  → MentorIntelligence (MI_LLM_BACKEND, MI_LLM_MODEL, MI_LLM_BASE_URL, MI_LLM_API_KEY)
+
+    Any variable left empty falls back to the global LLM_BACKEND / model / URL / key.
+
+    Example .env
+    ------------
+    # Route CodeIntelligence to a vLLM GPU server
+    CI_LLM_BACKEND=openai
+    CI_LLM_MODEL=codellama:34b
+    CI_LLM_BASE_URL=http://gpu-server:8000/v1
+
+    # Route DocIntelligence to a local Mistral (better multilingual)
+    DI_LLM_BACKEND=ollama
+    DI_LLM_MODEL=mistral:7b
+
+    # Route MentorIntelligence to Claude (best pedagogical quality)
+    MI_LLM_BACKEND=claude
+    MI_LLM_MODEL=claude-sonnet-4-5
+    """
+    from intelligence_core.config import settings
+
+    prefix = module.lower()  # "ci", "di", "mi"
+
+    backend  = getattr(settings, f"{prefix}_llm_backend",  "") or None
+    model    = getattr(settings, f"{prefix}_llm_model",    "") or None
+    base_url = getattr(settings, f"{prefix}_llm_base_url", "") or None
+    api_key  = getattr(settings, f"{prefix}_llm_api_key",  "") or None
+
+    if any([backend, model, base_url, api_key]):
+        logger.info(
+            "Module [%s] LLM override → backend=%s  model=%s  base_url=%s",
+            module.upper(),
+            backend  or "(global)",
+            model    or "(global)",
+            base_url or "(global)",
+        )
+
+    return get_llm_provider(backend, model=model, base_url=base_url, api_key=api_key)
+
+
+__all__ = [
+    "LLMProvider",
+    "SYSTEM_PROMPT_DEFAULT",
+    "get_llm_provider",
+    "get_module_llm_provider",
+]

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/intelligence_suite.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: intelligence-suite
-Version: 0.2.2
+Version: 0.2.4
 Summary: Modular knowledge retrieval suite: code, docs, APIs — all on-premise
 Author-email: ViciusLio <viciuslios@gmail.com>
 License-Expression: MIT
@@ -654,6 +654,28 @@ Switch backend with a single env var — no code changes required.
 > Any OpenAI-compatible server works with `LLM_BACKEND=openai` or `vllm` by pointing
 > `OPENAI_BASE_URL` at the correct endpoint.
 
+### Per-module LLM routing
+
+Each module can use a **different LLM backend and model** independently.
+Set any combination of `CI_LLM_*`, `DI_LLM_*`, `MI_LLM_*` in `.env` — leave empty to fall back to the global `LLM_BACKEND`:
+
+```env
+# CodeIntelligence → vLLM GPU server with code-specialised model
+CI_LLM_BACKEND=openai
+CI_LLM_MODEL=codellama:34b
+CI_LLM_BASE_URL=http://gpu-server:8000/v1
+
+# DocIntelligence → local Mistral (better multilingual / Italian)
+DI_LLM_BACKEND=ollama
+DI_LLM_MODEL=mistral:7b
+
+# MentorIntelligence → Claude API (best pedagogical quality)
+MI_LLM_BACKEND=claude
+MI_LLM_MODEL=claude-sonnet-4-5
+```
+
+Any OpenAI-compatible endpoint (vLLM, Groq, Mistral AI, LM Studio, Azure…) works by setting `*_LLM_BACKEND=openai` and `*_LLM_BASE_URL` to the endpoint.
+
 ### Escalation
 
 When retrieval confidence < `ESCALATION_THRESHOLD` and `ANTHROPIC_API_KEY` is set,

{intelligence_suite-0.2.2 → intelligence_suite-0.2.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "intelligence-suite"
-version = "0.2.2"
+version = "0.2.4"
 description = "Modular knowledge retrieval suite: code, docs, APIs — all on-premise"
 readme = "README.md"
 license = "MIT"

intelligence_suite-0.2.2/intelligence_core/llm/__init__.py

@@ -1,65 +0,0 @@
-"""LLM provider factory for Intelligence Suite.
-
-Supported backends
-------------------
-ollama       Local Ollama server (default, zero cost, no API key)
-openai       OpenAI Chat Completions API
-vllm         vLLM local GPU server (OpenAI-compatible)
-claude       Anthropic Claude API
-
-Any other OpenAI-compatible server (Groq, Mistral AI, LM Studio, Together AI, …)
-can be used with ``LLM_BACKEND=openai`` by changing ``OPENAI_BASE_URL``.
-
-Usage
------
-    from intelligence_core.llm import get_llm_provider
-
-    llm = get_llm_provider()          # reads LLM_BACKEND from .env
-    answer = llm.generate(question, context)
-"""
-
-from __future__ import annotations
-
-from intelligence_core.llm.protocol import LLMProvider, SYSTEM_PROMPT_DEFAULT
-
-
-def get_llm_provider(backend: str | None = None) -> LLMProvider:
-    """
-    Factory: return the configured LLM provider.
-
-    Args:
-        backend: Override ``LLM_BACKEND`` from settings.
-                 Values: ``"ollama"`` | ``"openai"`` | ``"vllm"`` | ``"claude"``
-
-    Returns:
-        An object satisfying the :class:`LLMProvider` protocol.
-    """
-    from intelligence_core.config import settings
-
-    _backend = backend or settings.llm_backend
-
-    if _backend in ("openai", "vllm"):
-        from intelligence_core.llm.openai_compat import OpenAICompatProvider
-        return OpenAICompatProvider(
-            base_url=settings.openai_base_url,
-            api_key=settings.openai_api_key,
-            model=settings.openai_model,
-            backend_hint=_backend,
-        )
-
-    if _backend == "claude":
-        from intelligence_core.llm.claude import ClaudeProvider
-        return ClaudeProvider(
-            api_key=settings.anthropic_api_key,
-            model=settings.claude_model,
-        )
-
-    # default: ollama
-    from intelligence_core.llm.ollama import OllamaProvider
-    return OllamaProvider(
-        base_url=settings.ollama_base_url,
-        model=settings.ollama_model,
-    )
-
-
-__all__ = ["LLMProvider", "SYSTEM_PROMPT_DEFAULT", "get_llm_provider"]