minder-cli 0.3.3__tar.gz → 0.3.4__tar.gz

{minder_cli-0.3.3 → minder_cli-0.3.4}/.gitignore

@@ -8,4 +8,5 @@ __pycache__
 .uv-cache
 ref
 .mcp.json
-.vscode/mcp.json
+.vscode/mcp.json
+default.profraw

{minder_cli-0.3.3 → minder_cli-0.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: minder-cli
-Version: 0.3.3
+Version: 0.3.4
 Summary: Minder CLI is the command-line interface for the Minder self-hosted MCP platform.
 Project-URL: Homepage, https://github.com/hiimtrung/minder
 Project-URL: Repository, https://github.com/hiimtrung/minder
@@ -19,8 +19,10 @@ Requires-Dist: fastapi>=0.136.0
 Requires-Dist: httpx>=0.28.0
 Provides-Extra: server
 Requires-Dist: aiosqlite>=0.21.0; extra == 'server'
+Requires-Dist: fastembed>=0.5.1; extra == 'server'
 Requires-Dist: langgraph>=1.1.8; extra == 'server'
 Requires-Dist: litellm>=1.83.1; extra == 'server'
+Requires-Dist: litert-lm-api-nightly>=0.10; extra == 'server'
 Requires-Dist: mcp>=1.26.0; extra == 'server'
 Requires-Dist: motor>=3.7.0; extra == 'server'
 Requires-Dist: passlib[bcrypt]>=1.7.4; extra == 'server'
@@ -35,86 +37,83 @@ Requires-Dist: zipp>=3.21.0; extra == 'server'
 Requires-Dist: zstandard>=0.25.0; extra == 'server'
 Description-Content-Type: text/markdown
 
-# Minder CLI
+# minder-cli
 
 [![PyPI version](https://img.shields.io/pypi/v/minder-cli.svg)](https://pypi.org/project/minder-cli/)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 
-**Minder CLI** is the command-line interface for the Minder platform—a self-hosted MCP (Model Context Protocol) platform for repository-aware engineering intelligence.
+**minder-cli** is the command-line interface for [Minder](https://github.com/hiimtrung/minder) — a self-hosted MCP platform for repository-aware engineering intelligence.
 
-## Features
-
-- **IDE Integration**: Scaffold repository-local MCP assets for VS Code, Cursor, and Claude Code.
-- **Repository Sync**: Auto-detect cross-repo dependencies and synchronize project knowledge.
-- **Self-Management**: Check for updates and upgrade both CLI and Server components in place.
-- **Unified Auth**: Simple login flow to connect your local development environment to your Minder Server.
+The CLI handles IDE scaffolding, repository sync, authentication, and self-updates. It connects to a **Minder Server** which runs the MCP gateway, RAG pipeline, memory engine, and admin console.
 
 ## Installation
 
-Install the CLI from PyPI using `uv` (recommended) or `pipx`:
-
 ```bash
+# Recommended
 uv tool install minder-cli
-# or
+
+# Alternative
 pipx install minder-cli
 ```
 
-## Quick Start
+## Requirements
 
-### 1. Connect to your Minder Server
+A running [Minder Server](https://github.com/hiimtrung/minder) — see the [server setup guide](https://github.com/hiimtrung/minder/blob/main/docs/minder-server.md) to get one running in minutes with Docker.
 
-Once you have a [Minder Server](docs/minder-server.md) running, log in with your client key:
+## Quick Start
+
+### 1. Connect to your server
 
 ```bash
 minder login --client-key mkc_your_client_key --server-url http://localhost:8800/sse
 ```
 
-### 2. Prepare your IDE
-
-Set up MCP tools and instructions for your favorite editor:
+### 2. Set up IDE integration
 
 ```bash
-# In your project root
+# In your project root — scaffolds MCP config for your editor
 minder install-ide --target vscode --target claude-code
 ```
 
-### 3. Sync Repository
+Supported targets: `vscode`, `cursor`, `claude-code`.
 
-Index your code and documentation to enable semantic search and RAG tools:
+### 3. Sync a repository
 
 ```bash
 minder sync --repo-id <repository-uuid>
 ```
 
-## Available Tool Surface
+Indexes code and documentation so AI agents can use semantic search and RAG tools.
 
-When connected, Minder provides powerful tools to your AI agents:
+## Commands
 
-| Tool | Description |
-| --- | --- |
-| `minder_query` | End-to-end RAG pipeline: retrieve + reason + verify |
-| `minder_search_code` | Semantic code retrieval across indexed repositories |
-| `minder_memory_recall` | Retrieve persisted memory entries |
-| `minder_workflow_get` | Read current workflow state |
+| Command | Description |
+|---------|-------------|
+| `minder login` | Authenticate the CLI against a Minder Server |
+| `minder install-ide` | Scaffold MCP assets for VS Code / Cursor / Claude Code |
+| `minder sync` | Index a repository into the Minder Server |
+| `minder check-update` | Check for available CLI and server updates |
+| `minder self-update` | Update CLI or server in place |
 
-## Maintenance
+## MCP Tools (once connected)
 
-Keep your tools up to date:
+Your AI agents will have access to:
 
-```bash
-minder check-update
-minder self-update --component cli
-minder self-update --component server
-```
+| Tool | Description |
+|------|-------------|
+| `minder_query` | Full RAG pipeline: retrieve → reason → verify → respond |
+| `minder_search_code` | Semantic code search across indexed repositories |
+| `minder_memory_recall` | Retrieve persisted engineering memory |
+| `minder_workflow_get` | Read current workflow state |
+| `minder_session_restore` | Restore session continuity across context windows |
 
-## Related Links
+## Links
 
-- [Development Workflow](docs/guides/development.md)
-- [Minder Server Setup](docs/minder-server.md)
-- [Admin & Client Onboarding](docs/guides/admin-client-onboarding.md)
-- [Local Setup Guide](docs/guides/local-setup.md)
-- [System Architecture](docs/system-design.md)
+- [GitHub Repository](https://github.com/hiimtrung/minder)
+- [Server Setup Guide](https://github.com/hiimtrung/minder/blob/main/docs/minder-server.md)
+- [Local Dev Setup](https://github.com/hiimtrung/minder/blob/main/docs/guides/local-setup.md)
+- [System Design](https://github.com/hiimtrung/minder/blob/main/docs/system-design.md)
 
 ## License
 
-Apache License 2.0. See [LICENSE](LICENSE) for details.
+Apache License 2.0. See [LICENSE](https://github.com/hiimtrung/minder/blob/main/LICENSE) for details.

minder_cli-0.3.4/README-pypi.md

@@ -0,0 +1,80 @@
+# minder-cli
+
+[![PyPI version](https://img.shields.io/pypi/v/minder-cli.svg)](https://pypi.org/project/minder-cli/)
+[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+
+**minder-cli** is the command-line interface for [Minder](https://github.com/hiimtrung/minder) — a self-hosted MCP platform for repository-aware engineering intelligence.
+
+The CLI handles IDE scaffolding, repository sync, authentication, and self-updates. It connects to a **Minder Server** which runs the MCP gateway, RAG pipeline, memory engine, and admin console.
+
+## Installation
+
+```bash
+# Recommended
+uv tool install minder-cli
+
+# Alternative
+pipx install minder-cli
+```
+
+## Requirements
+
+A running [Minder Server](https://github.com/hiimtrung/minder) — see the [server setup guide](https://github.com/hiimtrung/minder/blob/main/docs/minder-server.md) to get one running in minutes with Docker.
+
+## Quick Start
+
+### 1. Connect to your server
+
+```bash
+minder login --client-key mkc_your_client_key --server-url http://localhost:8800/sse
+```
+
+### 2. Set up IDE integration
+
+```bash
+# In your project root — scaffolds MCP config for your editor
+minder install-ide --target vscode --target claude-code
+```
+
+Supported targets: `vscode`, `cursor`, `claude-code`.
+
+### 3. Sync a repository
+
+```bash
+minder sync --repo-id <repository-uuid>
+```
+
+Indexes code and documentation so AI agents can use semantic search and RAG tools.
+
+## Commands
+
+| Command | Description |
+|---------|-------------|
+| `minder login` | Authenticate the CLI against a Minder Server |
+| `minder install-ide` | Scaffold MCP assets for VS Code / Cursor / Claude Code |
+| `minder sync` | Index a repository into the Minder Server |
+| `minder check-update` | Check for available CLI and server updates |
+| `minder self-update` | Update CLI or server in place |
+
+## MCP Tools (once connected)
+
+Your AI agents will have access to:
+
+| Tool | Description |
+|------|-------------|
+| `minder_query` | Full RAG pipeline: retrieve → reason → verify → respond |
+| `minder_search_code` | Semantic code search across indexed repositories |
+| `minder_memory_recall` | Retrieve persisted engineering memory |
+| `minder_workflow_get` | Read current workflow state |
+| `minder_session_restore` | Restore session continuity across context windows |
+
+## Links
+
+- [GitHub Repository](https://github.com/hiimtrung/minder)
+- [Server Setup Guide](https://github.com/hiimtrung/minder/blob/main/docs/minder-server.md)
+- [Local Dev Setup](https://github.com/hiimtrung/minder/blob/main/docs/guides/local-setup.md)
+- [System Design](https://github.com/hiimtrung/minder/blob/main/docs/system-design.md)
+
+## License
+
+Apache License 2.0. See [LICENSE](https://github.com/hiimtrung/minder/blob/main/LICENSE) for details.

minder_cli-0.3.4/README.md

@@ -0,0 +1,96 @@
+# Minder
+
+[![PyPI version](https://img.shields.io/pypi/v/minder-cli.svg)](https://pypi.org/project/minder-cli/)
+[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+
+**Minder** is a self-hosted MCP (Model Context Protocol) platform for repository-aware engineering intelligence.
+
+It combines a local-first inference stack, a persistent memory and workflow engine, and a developer-facing CLI into a single deployable unit.
+
+## What's in this repo
+
+| Component | Description |
+|-----------|-------------|
+| **Minder Server** | MCP gateway — SSE + stdio transport, RAG pipeline, workflow engine, memory, admin HTTP |
+| **Minder CLI** (`minder-cli` on PyPI) | Edge CLI — IDE scaffold, repo sync, login, self-update |
+| **Minder Dashboard** | Astro admin console — client management, onboarding, skill catalog |
+
+## Architecture
+
+```
+Developer → minder-cli → Minder Server ←→ AI agents (Codex / Copilot / Claude)
+                              │
+               ┌──────────────┼──────────────┐
+               │              │              │
+           MongoDB          Redis         Milvus
+         (graph/memory)   (cache)      (vector search)
+               │
+          ┌─────┴──────┐
+          │            │
+       LiteRT-LM    FastEmbed
+       (LLM gen)   (embedding)
+```
+
+- **LLM inference**: LiteRT-LM (Google AI Edge) — on-device, hardware-accelerated, no HTTP overhead
+- **Embedding inference**: FastEmbed running natively in-process (`mxbai-embed-large-v1`) — zero-dependency ONNX acceleration
+
+## Quick Start
+
+### Run the server
+
+```bash
+# 1. Download the LiteRT-LM model
+./scripts/download_models.sh
+
+# 2. Start infra (MongoDB + Redis + Milvus)
+docker compose -f docker/docker-compose.local.yml up -d
+
+# 3. Run Minder Server
+uv run python -m minder.server
+```
+
+Or use the one-command release installer (Docker only, no local uv needed):
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/hiimtrung/minder/main/scripts/release/install-minder-release.sh | bash
+```
+
+### Connect the CLI
+
+```bash
+# Install
+uv tool install minder-cli
+
+# Log in with your client key
+minder login --client-key mkc_your_key --server-url http://localhost:8800/sse
+
+# Set up IDE integration (VS Code, Cursor, Claude Code)
+minder install-ide --target vscode --target claude-code
+
+# Sync a repository
+minder sync --repo-id <uuid>
+```
+
+## MCP Tools
+
+When connected, Minder exposes these tools to your AI agents:
+
+| Tool | Description |
+|------|-------------|
+| `minder_query` | Full RAG pipeline: retrieve → reason → verify → respond |
+| `minder_search_code` | Semantic code search across indexed repos |
+| `minder_memory_recall` | Retrieve persisted engineering memory |
+| `minder_workflow_get` | Read current workflow state |
+| `minder_session_restore` | Restore session continuity across context windows |
+
+## Documentation
+
+- [System Design](docs/system-design.md)
+- [Server Setup](docs/minder-server.md)
+- [Local Dev Setup](docs/guides/local-setup.md)
+- [Production Deployment](docs/guides/production-deployment.md)
+- [Admin & Client Onboarding](docs/guides/admin-client-onboarding.md)
+
+## License
+
+Apache License 2.0. See [LICENSE](LICENSE) for details.

{minder_cli-0.3.3 → minder_cli-0.3.4}/pyproject.toml

@@ -4,9 +4,9 @@ build-backend = "hatchling.build"
 
 [project]
 name = "minder-cli"
-version = "0.3.3"
+version = "0.3.4"
 description = "Minder CLI is the command-line interface for the Minder self-hosted MCP platform."
-readme = "README.md"
+readme = "README-pypi.md"
 requires-python = ">=3.14"
 keywords = ["mcp", "cli", "ai", "code-search", "workflow"]
 classifiers = [
@@ -26,8 +26,10 @@ dependencies = [
 [project.optional-dependencies]
 server = [
     "aiosqlite>=0.21.0",
+    "fastembed>=0.5.1",
     "langgraph>=1.1.8",
     "litellm>=1.83.1",
+    "litert-lm-api-nightly>=0.10",
     "mcp>=1.26.0",
     "motor>=3.7.0",
     "passlib[bcrypt]>=1.7.4",
@@ -67,6 +69,7 @@ packages = ["src/minder"]
 [tool.hatch.build.targets.sdist]
 include = [
     "/README.md",
+    "/README-pypi.md",
     "/pyproject.toml",
     "/src/minder",
 ]

minder_cli-0.3.4/src/minder/cli.py

@@ -0,0 +1,7 @@
+from __future__ import annotations
+
+import sys
+from minder.presentation.cli.main import main
+
+if __name__ == "__main__":
+    sys.exit(main())

{minder_cli-0.3.3 → minder_cli-0.3.4}/src/minder/config.py

@@ -31,18 +31,21 @@ class AuthConfig(BaseModel):
 
 
 class EmbeddingConfig(BaseModel):
-    provider: str = "ollama"
-    ollama_url: str = "http://localhost:11434"
-    ollama_model: str = "qwen3-embedding:0.6b"
-    dimensions: int = 768
+    provider: str = "fastembed"
+    runtime: str = "auto"  # "auto" | "fastembed" | "mock"
+    fastembed_model: str = "mixedbread-ai/mxbai-embed-large-v1"
+    fastembed_cache_dir: str = "~/.minder/cache/fastembed"
+    dimensions: int = 1024
     openai_api_key: Optional[str] = None
     openai_model: str = "text-embedding-3-small"
 
 
 class LLMConfig(BaseModel):
-    provider: str = "ollama"
-    ollama_url: str = "http://localhost:11434"
-    ollama_model: str = "qwen3.5:4b"
+    provider: str = "litert"  # "litert" | "openai"
+    # LiteRT-LM fields
+    litert_model_path: str = "~/.minder/models/gemma-4-E2B-it.litertlm"
+    litert_backend: str = "auto"  # "auto" (GPU on Mac, CPU elsewhere) | "cpu" | "gpu"
+    litert_cache_dir: str = "~/.minder/cache/litert"
     context_length: int = 131072
     temperature: float = 0.1
     openai_api_key: Optional[str] = None

{minder_cli-0.3.3 → minder_cli-0.3.4}/src/minder/continuity.py

@@ -268,14 +268,10 @@ def _extract_json_object(raw: str) -> dict[str, Any] | None:
 
 class ContinuitySynthesizer:
     def __init__(self, config: MinderConfig) -> None:
-        from minder.llm.local import LocalModelLLM
+        from minder.llm.factory import create_llm
 
         self._config = config
-        self._llm = LocalModelLLM(
-            ollama_url=config.llm.ollama_url,
-            ollama_model=config.llm.ollama_model,
-            context_length=config.llm.context_length,
-        )
+        self._llm = create_llm(config.llm)
 
     def synthesize_memory_hits(
         self,
@@ -304,7 +300,7 @@ class ContinuitySynthesizer:
         )
         raw = self._llm.complete_text(
             prompt,
-            max_tokens=700,
+            max_tokens=1000,
             temperature=min(max(self._config.llm.temperature, 0.05), 0.3),
             fallback="",
         )
@@ -312,7 +308,7 @@ class ContinuitySynthesizer:
         if not parsed:
             return fallback, {
                 "provider": "heuristic",
-                "model": self._config.llm.ollama_model,
+                "model": self._config.llm.provider,
                 "runtime": self._llm.runtime,
             }
         return {
@@ -329,8 +325,8 @@ class ContinuitySynthesizer:
                 ).items()
             },
         }, {
-            "provider": "local_llm",
-            "model": self._config.llm.ollama_model,
+            "provider": "litert_lm",
+            "model": self._config.llm.provider,
             "runtime": self._llm.runtime,
         }
 

minder_cli-0.3.4/src/minder/embedding/local.py

@@ -0,0 +1,184 @@
+"""
+Local Embedding provider — delegates to FastEmbed using ONNX runtime.
+
+Falls back to a deterministic hash-based stub if initialization fails.
+"""
+
+from __future__ import annotations
+
+import gc
+import hashlib
+import logging
+import math
+from collections import OrderedDict
+from pathlib import Path
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+
+_MODEL_CACHE: dict[str, Any] = {}
+_EMBEDDING_CACHE: OrderedDict[str, list[float]] = OrderedDict()
+MAX_CACHE_SIZE = 100
+MAX_TEXT_LENGTH = 8000  # Safety truncation to avoid over-context (~2000 tokens)
+
+
+class LocalEmbeddingProvider:
+    def __init__(
+        self,
+        fastembed_model: str = "mixedbread-ai/mxbai-embed-large-v1",
+        fastembed_cache_dir: str = "~/.minder/cache/fastembed",
+        dimensions: int = 1024,
+        runtime: str = "auto",
+    ) -> None:
+        self._model_name = fastembed_model
+        self._cache_dir = str(Path(fastembed_cache_dir).expanduser())
+        self._dimensions = dimensions
+        self._runtime = runtime
+        self._model: Any | None = None
+        self._init_model()
+
+    def _init_model(self) -> None:
+        if self._runtime == "mock":
+            return
+
+        cache_key = f"{self._model_name}:{self._cache_dir}"
+        if cache_key in _MODEL_CACHE:
+            self._model = _MODEL_CACHE[cache_key]
+            return
+
+        try:
+            from fastembed import TextEmbedding  # type: ignore[import-not-found]
+
+            # Optimize for speed and resource usage:
+            # - threads=4 limits CPU usage while maintaining good throughput
+            # - lazy_load=False ensures first request is fast
+            self._model = TextEmbedding(
+                model_name=self._model_name,
+                cache_dir=self._cache_dir,
+                threads=4,
+            )
+            _MODEL_CACHE[cache_key] = self._model
+        except Exception as e:
+            logger.warning(
+                f"Failed to initialize FastEmbed model {self._model_name}: {e}. Using mock."
+            )
+            self._model = None
+
+    @property
+    def runtime(self) -> str:
+        if self._runtime != "auto":
+            return self._runtime
+        return "fastembed" if self._model is not None else "mock"
+
+    def embed(self, text: str) -> list[float]:
+        if not text:
+            return [0.0] * self._dimensions
+
+        # 1. Truncate to avoid over-context errors
+        safe_text = text[:MAX_TEXT_LENGTH]
+        
+        # 2. Check cache
+        if safe_text in _EMBEDDING_CACHE:
+            _EMBEDDING_CACHE.move_to_end(safe_text)
+            return _EMBEDDING_CACHE[safe_text]
+
+        # 3. Perform embedding
+        embedding: list[float]
+        if self.runtime == "fastembed" and self._model is not None:
+            try:
+                # FastEmbed returns a generator of numpy arrays
+                embeddings = list(self._model.embed([safe_text]))
+                if embeddings:
+                    embedding = embeddings[0].tolist()[: self._dimensions]
+                else:
+                    embedding = self._hash_embed(safe_text)
+            except Exception as e:
+                logger.warning(f"FastEmbed failed during inference: {e}")
+                embedding = self._hash_embed(safe_text)
+        else:
+            embedding = self._hash_embed(safe_text)
+
+        # 4. Update cache (LRU)
+        _EMBEDDING_CACHE[safe_text] = embedding
+        if len(_EMBEDDING_CACHE) > MAX_CACHE_SIZE:
+            _EMBEDDING_CACHE.popitem(last=False)
+
+        return embedding
+
+    def embed_many(self, texts: list[str]) -> list[list[float]]:
+        if not texts:
+            return []
+
+        results: list[list[float]] = [[] for _ in texts]
+        to_embed_indices: list[int] = []
+        to_embed_texts: list[str] = []
+
+        # 1. Try cache first
+        for i, text in enumerate(texts):
+            safe_text = text[:MAX_TEXT_LENGTH] if text else ""
+            if not safe_text:
+                results[i] = [0.0] * self._dimensions
+            elif safe_text in _EMBEDDING_CACHE:
+                _EMBEDDING_CACHE.move_to_end(safe_text)
+                results[i] = _EMBEDDING_CACHE[safe_text]
+            else:
+                to_embed_indices.append(i)
+                to_embed_texts.append(safe_text)
+
+        if not to_embed_texts:
+            return results
+
+        # 2. Batch embed the missing ones
+        if self.runtime == "fastembed" and self._model is not None:
+            try:
+                embeddings = list(self._model.embed(to_embed_texts))
+                for i, emb in enumerate(embeddings):
+                    idx = to_embed_indices[i]
+                    vector = emb.tolist()[: self._dimensions]
+                    results[idx] = vector
+                    # Update cache
+                    _EMBEDDING_CACHE[to_embed_texts[i]] = vector
+            except Exception as e:
+                logger.warning(f"FastEmbed batch failed: {e}")
+                for i, idx in enumerate(to_embed_indices):
+                    vector = self._hash_embed(to_embed_texts[i])
+                    results[idx] = vector
+                    _EMBEDDING_CACHE[to_embed_texts[i]] = vector
+        else:
+            for i, idx in enumerate(to_embed_indices):
+                vector = self._hash_embed(to_embed_texts[i])
+                results[idx] = vector
+                _EMBEDDING_CACHE[to_embed_texts[i]] = vector
+
+        # 3. Cache maintenance
+        while len(_EMBEDDING_CACHE) > MAX_CACHE_SIZE:
+            _EMBEDDING_CACHE.popitem(last=False)
+
+        return results
+
+    def _hash_embed(self, text: str) -> list[float]:
+        """Generate a deterministic but word-aware mock embedding."""
+        words = text.lower().split()
+        if not words:
+            return [0.0] * self._dimensions
+
+        vector = [0.0] * self._dimensions
+        for word in words:
+            h = hashlib.sha256(word.encode()).digest()
+            for i in range(self._dimensions):
+                byte_val = h[i % len(h)]
+                vector[i] += (byte_val / 255.0) - 0.5
+
+        norm = math.sqrt(sum(v * v for v in vector))
+        if norm > 1e-9:
+            return [v / norm for v in vector]
+        return [0.0] * self._dimensions
+
+def clear_caches() -> None:
+    """Clear global model and embedding caches to reclaim memory."""
+    global _MODEL_CACHE, _EMBEDDING_CACHE
+    _MODEL_CACHE.clear()
+    _EMBEDDING_CACHE.clear()
+    gc.collect()
+    logger.debug("Cleared FastEmbed global caches.")

{minder_cli-0.3.3 → minder_cli-0.3.4}/src/minder/graph/graph.py

@@ -23,7 +23,7 @@ from minder.graph.nodes import (
     WorkflowPlannerNode,
 )
 from minder.graph.state import GraphState
-from minder.llm.local import LocalModelLLM
+from minder.llm.factory import create_llm
 from minder.llm.openai import OpenAIFallbackLLM
 from minder.store.interfaces import (
     IOperationalStore,
@@ -58,8 +58,8 @@ class MinderGraph:
         self._planning = planning or PlanningNode()
         vector_store = VectorStore(store, store)
         embedder = LocalEmbeddingProvider(
-            ollama_url=config.embedding.ollama_url,
-            ollama_model=config.embedding.ollama_model,
+            fastembed_model=config.embedding.fastembed_model,
+            fastembed_cache_dir=config.embedding.fastembed_cache_dir,
             dimensions=config.embedding.dimensions,
             runtime="auto",
         )
@@ -72,11 +72,7 @@ class MinderGraph:
         self._reranker = reranker  # None by default; pass RerankerNode(...) to activate
         self._reasoning = reasoning or ReasoningNode()
         self._llm = llm or LLMNode(
-            primary=LocalModelLLM(
-                ollama_url=config.llm.ollama_url,
-                ollama_model=config.llm.ollama_model,
-                context_length=config.llm.context_length,
-            ),
+            primary=create_llm(config.llm),
             fallback=OpenAIFallbackLLM(
                 config.llm.openai_api_key,
                 config.llm.openai_model,

minder_cli-0.3.4/src/minder/llm/__init__.py

@@ -0,0 +1,11 @@
+from .base import LLMClient
+from .factory import create_llm
+from .litert import LiteRTModelLLM
+from .openai import OpenAIFallbackLLM
+
+__all__ = [
+    "LLMClient",
+    "LiteRTModelLLM",
+    "OpenAIFallbackLLM",
+    "create_llm",
+]