Install¶

Riptide-Watergraph is on PyPI. The core is dependency-light; heavier integrations are optional extras imported lazily.

Core¶

pip install riptide-watergraph

This gives you the graph engine, the 238-tool registry, the 219-role catalog, memory, guardrails, the swarm composer, the eval harness, and the riptide CLI — all runnable offline with the deterministic DemoGateway.

Extras¶

Extra	Installs	For
`[server]`	FastAPI + uvicorn	`riptide serve` + the Like Water Studio web UI
`[litellm]`	litellm	running against a real LLM (any OpenAI-compatible provider)
`[mcp]`	mcp SDK	the real MCP stdio transport (`StdioMcpClient`)
`[observability]`	OpenTelemetry + Langfuse	per-LLM-call tracing spans
`[pgvector]`	langchain-postgres + psycopg	the Postgres/pgvector memory backend
`[all]`	litellm + observability + mcp + server	everything for a real-model, traced, served run
`[dev]`	pytest, ruff, mypy, httpx	contributing

pip install "riptide-watergraph[server]"     # web UI
pip install "riptide-watergraph[all]"        # the lot

From source¶

git clone https://github.com/shibinsp/riptide-watergraph.git
cd riptide-watergraph
pip install -e ".[dev,server]"

Verify¶

riptide run "What is 21 * 2?" --offline
riptide eval --offline          # 4/4 = 100%

Docker¶

docker build -t riptide-watergraph .
docker run -p 8000:8000 riptide-watergraph     # then open http://localhost:8000/

See Deploy for Docker Compose and hosting notes.