Installation¶
The base install gets you the Architecture base class, every architecture's Python class, and the LangGraph/LangChain runtime. For LLM providers, vector stores, and tools you pick the extras you want:
Then set NEBIUS_API_KEY in .env. Pre-configured to use meta-llama/Llama-3.3-70B-Instruct for general work and Qwen/Qwen3-235B-A22B-Thinking-2507-fast for reasoning architectures (nb 19, 20, 21, 22).
Set OPENAI_API_KEY and LLM_PROVIDER=openai (with LLM_MODEL=gpt-4o-mini or similar).
Set ANTHROPIC_API_KEY and LLM_PROVIDER=anthropic.
Run ollama pull llama3.3:70b first. Set LLM_PROVIDER=ollama. No API key needed.
Memory backend extras¶
| Extra | What it gets you | Use when |
|---|---|---|
faiss (recommended) |
FAISS-CPU + langchain-community | Local in-process vector store. Used by nb 08, 18, 23-26, 29, 31, 35. |
chroma |
Chroma + chromadb | When you want persistence and an HTTP server. |
qdrant |
Qdrant + qdrant-client | Production-grade, persistent, multi-tenant. |
neo4j |
langchain-neo4j + neo4j driver | For nb 08, 12, 27 — graph queries via Cypher. |
networkx |
NetworkX (in-process graph) | Default graph backend; no extra service to run. |
Tool extras¶
| Extra | Used by |
|---|---|
tavily |
Web search — nb 02, 03, 04, 05, 07, 11, 24, 30 |
Browser-using architecture (nb 34)¶
Notebook 34 (BrowserAgent) drives a real headless Chromium. After the base install:
playwright install chromium downloads ~200 MB of browser binaries. Run once per machine.
Verify your install¶
python -c "from agentic_architectures import get_llm, settings; print(settings.llm_provider, settings.llm_model)"
# nebius meta-llama/Llama-3.3-70B-Instruct
If you see your configured provider and model, you're ready. Continue to the Quickstart.