FlowFoundry

FlowFoundry is a cloud-agnostic agentic workflow framework built on LangGraph and LangChain.
It helps you design, test, and run agentic workflows locally or in the cloud, with pluggable connectors for storage, LLMs, rerankers, and external tools.

Features

• 🔌 Cloud-agnostic core
• 🧠 Multi-LLM per node
• 🧪 Testable with in-memory components
• 🛠️ Extensible for RAG, tool use, DB retrieval, form filing, image interpretation

📦 Install

python -m venv .venv && source .venv/bin/activate
pip install -e ".[rag,search,rerank,qdrant,openai,llm-openai,dev]"

Examples run offline with echo LLM. Optional extras no-op gracefully.

Repository layout

flowfoundry/
├─ examples/                  # ready-to-run Python demos
├─ src/flowfoundry/
│  ├─ functional/             # core strategy functions (chunk, index, rerank, compose)
│  ├─ model/                  # LLM provider wrappers (openai, ollama, hf, langchain)
│  ├─ utils/                  # registry, errors, helpers
│  └─ cli.py                  # Typer CLI (auto-discovers functional strategies)
├─ tests/                     # pytest functional tests
├─ pyproject.toml
└─ Makefile

Concepts

• Strategy: pure function (e.g. chunk_recursive)

• Functional API: call strategies directly

• Blocks: wrappers with config (Recursive, ChromaUpsert, etc.)

• Node: registered stateful step in LangGraph

• Workflow: nodes + edges compiled into a runnable graph

Examples

Functional API

from flowfoundry.functional import (
    chunk_recursive,
    index_chroma_upsert,
    index_chroma_query,
)

chunks = chunk_recursive("FlowFoundry example text", chunk_size=120, chunk_overlap=20, doc_id="demo")
index_chroma_upsert(chunks, path=".ff_chroma", collection="docs")
hits = index_chroma_query("What is FlowFoundry?", path=".ff_chroma", collection="docs", k=5)

CLI

Every registered function is available automatically:

# list all available strategies
flowfoundry list

# run a strategy by family/name
flowfoundry call chunking fixed --kwargs '{"data":"hello world","chunk_size":5}'

Or use subcommands (auto-generated per family):

flowfoundry chunking fixed --kwargs '{"data":"hello world","chunk_size":5}'
flowfoundry indexing chroma_query --kwargs '{"q":"budget","path":".ff_chroma","collection":"docs"}'

Ready-to-run Python scripts are in examples/python/:

01_load_chunk_index.py – load PDFs, chunk, index into Chroma

02_query_rerank.py – query and rerank hits

03_compose_openai.py – retrieve + answer with OpenAI

04_compose_ollama.py – same but using local Ollama

06_pipeline_end_to_end.py – full pipeline in one script

Run:

python examples/python/01_load_chunk_index.py

YAML Plans (no-code pipelines)

Run FlowFoundry strategies from a declarative YAML “plan”—no Python needed.

# basic usage
flowfoundry run path/to/plan.yaml

# debug: print intermediate step outputs too
flowfoundry run path/to/plan.yaml --print-steps

Example Script

flowfoundry run examples/yaml/rag_sample.yaml

Development

make dev      # editable install + extras
make test     # run pytest
make docs     # build Sphinx docs

Pre-commit hooks:

pre-commit install
pre-commit run --all-files

Contributing

• Fork the repo

• Create a branch: git checkout -b feat/my-feature

• Commit + push

• Open a PR

Docs

Build locally:

cd docs && make html
open _build/html/index.html