nodex-ai 0.1.0

The Express.js-style developer experience for LangGraph agents.

nodex

The Express.js-style developer experience for LangGraph agents.

Nodex is a thin layer on top of LangGraph that lets you build agents with decorators, middleware, retries, fallbacks, tracing, and a small CLI.

It does not replace LangGraph. It removes the boilerplate around it.

Why

LangGraph is powerful, but even a small agent can require manual state types, node registration, edge wiring, retry handling, and debugging through several layers of framework internals.

Nodex gives that workflow a smaller surface:

• Define graph steps with @app.node.
• Add cross-cutting behavior with @app.middleware.
• Route conditionally with @app.route.
• Run the graph with app.run(...).
• Get a terminal execution trace automatically.

Install

pip install nodex-ai

Requires Python 3.11 or newer.

Quickstart

from nodex import Agent

app = Agent(name="hello-agent")


@app.node(next="writer")
def research(state):
    query = state.get("query", "LangGraph")
    return {"notes": f"Research notes about {query}"}


@app.node(next="end")
def writer(state):
    return {"final": state.get("notes").upper()}


if __name__ == "__main__":
    app.run({"query": "agent frameworks"})

Output:

>  nodex running - hello-agent
--------------------------------------------------
OK  research       ->  0.002s  |  tokens: 0  |  $0.0000
OK  writer         ->  0.001s  |  tokens: 0  |  $0.0000
--------------------------------------------------
OK  Completed  |  0.023s  |  $0.0000  |  2 nodes  |  0 failed  |  tokens: 0

LangGraph vs Nodex

LangGraph:

from langgraph.graph import END, START, StateGraph

graph = StateGraph(dict)
graph.add_node("research", research)
graph.add_node("writer", writer)
graph.add_edge(START, "research")
graph.add_edge("research", "writer")
graph.add_edge("writer", END)
app = graph.compile()

Nodex:

from nodex import Agent

app = Agent()


@app.node(next="writer")
def research(state):
    return {"notes": "research"}


@app.node(next="end")
def writer(state):
    return {"final": state.get("notes")}


app.run()

Core API

Agent

from nodex import Agent

app = Agent(name="research-agent", debug=True)

Agent owns node registration, middleware registration, graph compilation, and execution tracing.

Nodes

@app.node(next="writer", retry=2, on_fail="fallback_research")
def research(state):
    return {"research_results": "data"}

Node functions receive a NodexState object and must return a non-empty dictionary. Internal keys such as _trace and _retry_count are protected.

Middleware

@app.middleware
def logger(state, next_node):
    print(f"running {state._current_node}")
    return next_node(state)

Middleware runs in registration order and wraps node execution.

Conditional routing

@app.node()
@app.route(
    condition=lambda state: state.get("confidence", 0) > 0.8,
    if_true="publish",
    if_false="review",
)
def writer(state):
    return {"draft": "ready"}

Human review

@app.node(next="publish", human_review=True)
def review(state):
    return {"approved_content": state.get("draft")}

When human_review=True, Nodex asks for terminal approval before continuing.

CLI

nodex new my-agent
cd my-agent
nodex dev agent:app
nodex run agent:app --input '{"query":"hello"}'

Testing nodes

from nodex.testing import assert_node_output, test_node


def test_research():
    result = test_node(research, {"query": "AI trends"})
    assert result.success
    assert "research_results" in result.output


def test_required_keys():
    assert_node_output(research, {"query": "AI trends"}, ["research_results"])

Local development

pip install -e ".[dev]"
ruff check src tests
pytest

Current test suite: 58 tests, with app, graph, CLI, decorators, middleware, state, tracer, errors, and testing helpers covered.

Documentation

The local docs site lives in site/ and can be served with:

npm run dev

Then open http://127.0.0.1:4173/.

Roadmap

• Decorator-based node definition
• Middleware engine
• Retry and fallback handling
• Terminal execution tracing
• CLI scaffold/run/dev commands
• Testing helpers
• Multi-page docs site
• Provider-specific examples
• Dashboard or trace viewer

License

MIT. See LICENSE.