feat: introduce unified LLM client for multi-provider support and add project requirements document for v2 init build

Author: jenilsavani9

packages/cli/agents/domain_extractor.py

@@ -26,8 +26,7 @@
 from pathlib import Path
 from typing import Any
 
-from packages.arc42gen.models.config import LLMConfig
-from packages.arc42gen.providers.factory import create_llm_provider
+from packages.cli.agents.llm_client import client_from_settings
 from packages.cli.graph.neo4j_client import NeoClient
 from packages.config.settings import Settings
 
@@ -83,41 +82,6 @@
 _COUNT_EDGES = "MATCH ()-[r:IMPLEMENTS_DOMAIN]->() RETURN count(r) AS c"
 
 
-# ---------------------------------------------------------------------------
-# LLM config builder (mirrors summarizer pattern)
-# ---------------------------------------------------------------------------
-
-def _build_llm_config(settings: Settings) -> LLMConfig:
-    provider = settings.LLM_PROVIDER.lower()
-    if provider == "ollama":
-        return LLMConfig(
-            provider="ollama",
-            model=settings.LLM_MODEL_OLLAMA,
-            api_key="",
-            base_url=settings.OLLAMA_BASE_URL,
-            max_tokens=2000,
-        )
-    elif provider == "anthropic":
-        return LLMConfig(
-            provider="anthropic",
-            model=LLMConfig.DEFAULT_MODELS["anthropic"],
-            api_key=settings.ANTHROPIC_API_KEY,
-            max_tokens=2000,
-        )
-    elif provider == "gemini":
-        return LLMConfig(
-            provider="gemini",
-            model=LLMConfig.DEFAULT_MODELS["gemini"],
-            api_key=settings.GEMINI_API_KEY,
-            max_tokens=2000,
-        )
-    else:
-        raise ValueError(
-            f"Unsupported LLM_PROVIDER '{provider}'. "
-            "Set LLM_PROVIDER to ollama, anthropic, or gemini in .env"
-        )
-
-
 # ---------------------------------------------------------------------------
 # Prompt builder
 # ---------------------------------------------------------------------------
@@ -258,8 +222,7 @@ def run_domain_extractor(
         progress_cb("fetch", f"Loaded {len(nodes)} nodes.")
 
     # 2. Call LLM in batches; merge domains across responses
-    llm_config = _build_llm_config(settings)
-    llm = create_llm_provider(llm_config)
+    llm = client_from_settings(settings)
 
     merged: dict[str, dict] = {}
     batches = [nodes[i : i + batch_size] for i in range(0, len(nodes), batch_size)]
@@ -269,8 +232,8 @@ def run_domain_extractor(
             progress_cb("llm", f"LLM call {i}/{len(batches)} ({len(batch)} nodes)...")
         prompt = _build_batch_prompt(batch)
         try:
-            response = llm.generate(prompt=prompt, temperature=0.1)
-            parsed   = _extract_json(response.content)
+            raw_text = llm.complete(prompt, max_tokens=2000, temperature=0.1)
+            parsed   = _extract_json(raw_text)
             domains  = parsed.get("domains", [])
             _merge_domains(merged, domains)
         except Exception as exc:

packages/cli/agents/embedder.py

@@ -2,21 +2,23 @@
 Embedding agent.
 
 Fetches nodes that have a summary but no embedding from Neo4j,
-generates a vector via Ollama, and writes it back.
+generates a vector via the configured LLM client, and writes it back.
 
 Also creates Neo4j vector indexes on first run.
 
 Config from packages.config.settings.Settings:
-    OLLAMA_BASE_URL         Ollama host (default http://localhost:11434)
-    OLLAMA_EMBEDDING_MODEL  Model for embeddings (default mxbai-embed-large)
-    EMBEDDING_DIMENSION     Vector dimensions (must match the model, default 1024)
+    LLM_PROVIDER            ollama | openai | anthropic
+    OLLAMA_BASE_URL         Ollama host  (ollama + anthropic embed back-end)
+    OLLAMA_EMBEDDING_MODEL  Embed model  (ollama + anthropic)
+    OPENAI_EMBEDDING_MODEL  Embed model  (openai)
+    OPENAI_API_KEY          Required when LLM_PROVIDER=openai
+    EMBEDDING_DIMENSION     Vector dimensions (must match the model)
 """
 from __future__ import annotations
 
 from typing import Any
 
-import requests
-
+from packages.cli.agents.llm_client import client_from_settings
 from packages.cli.graph.neo4j_client import NeoClient
 from packages.config.settings import Settings
 
@@ -56,22 +58,6 @@ def ensure_vector_indexes(client: NeoClient, dimensions: int) -> None:
 """
 
 
-# ---------------------------------------------------------------------------
-# Ollama embed call
-# ---------------------------------------------------------------------------
-
-def _embed(text: str, settings: Settings) -> list[float]:
-    """Call Ollama /api/embeddings and return the embedding vector."""
-    url = f"{settings.OLLAMA_BASE_URL.rstrip('/')}/api/embeddings"
-    resp = requests.post(
-        url,
-        json={"model": settings.OLLAMA_EMBEDDING_MODEL, "prompt": text},
-        timeout=60,
-    )
-    resp.raise_for_status()
-    return resp.json()["embedding"]
-
-
 # ---------------------------------------------------------------------------
 # Core runner
 # ---------------------------------------------------------------------------
@@ -96,6 +82,7 @@ def run_embedder(
     """
     ensure_vector_indexes(client, settings.EMBEDDING_DIMENSION)
 
+    llm = client_from_settings(settings)
     counts: dict[str, int] = {label: 0 for label in _LABELS}
 
     def _write(tx: Any, node_id: str, embedding: list[float]) -> None:
@@ -111,7 +98,7 @@ def _write(tx: Any, node_id: str, embedding: list[float]) -> None:
 
             for node in rows:
                 try:
-                    embedding = _embed(node["summary"], settings)
+                    embedding = llm.embed(node["summary"])
                 except Exception as exc:
                     print(f"    [warn] embedding failed for {node['id']}: {exc}")
                     continue