Dependencies and Extras#
Core Install#
pip install design-research-agents
Editable contributor setup:
git clone https://github.com/cmudrc/design-research-agents.git
cd design-research-agents
python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -e ".[dev]"
Or use:
make dev
Maintainer release baseline#
Use this when preparing a tagged release:
Use Python
3.12(from.python-version).Install maintainer dependencies:
make dev.Verify full checks:
make ci.Build release artifacts and validate metadata:
make release-check.Commit dependency spec changes, then tag and publish.
make release-check builds both the source distribution and wheel into dist/
and runs twine check against the generated artifacts.
Extras matrix#
Extra |
Purpose |
|---|---|
|
Contributor tooling |
|
OpenAI-family hosted SDK backends |
|
Anthropic hosted backend |
|
Gemini hosted backend |
|
Groq hosted backend |
|
MCP tool-runtime integration |
|
Hugging Face Hub metadata discovery for |
|
Optional ChromaDB-backed vector memory store |
|
Optional NetworkX-backed graph memory store |
|
Managed llama.cpp backend |
|
In-process transformers backend |
|
Apple MLX backend |
|
vLLM server backend (Linux) |
|
SGLang server backend (Linux) |
|
Local-backend convenience bundle |
|
Hosted-provider convenience bundle |
|
Providers + local backends |
|
|
full remains the backend-focused bundle. Use all when you want that same
runtime surface plus the optional memory backends exposed by this package.
Hosted clients are the fastest path for onboarding and benchmark iteration, but they require network access and data egress. Local in-process clients are often preferable for privacy-sensitive studies and single-machine experimentation, but they are more hardware-sensitive. Server-backed local clients improve deployment flexibility and throughput isolation, but they add service-management overhead.
Recommended install profiles:
hosted OpenAI-family studies:
pip install -e ".[dev,openai]"hosted provider comparisons:
pip install -e ".[dev,providers]"Hugging Face catalog discovery:
pip install -e ".[dev,huggingface]"Chroma-backed memory experiments:
pip install -e ".[dev,memory_chroma]"graph-memory experiments:
pip install -e ".[dev,memory_graph]"local-only studies:
pip install -e ".[dev,local]"broad backend validation:
pip install -e ".[dev,full]"broad runtime + memory validation:
pip install -e ".[dev,all]"
Release validation is exposed via make release-check.