Build apps in one primitive that spans your whole stack — databases, UIs, AI agents, and services.
The primitive Nu is built on, in three words.
A tiny script keeps its values in memory and calls functions on them. Real apps don't stay there. Values live in a database, arrive from a browser form, render into a dashboard, get recomputed by a background job. Almost none of the code is about the values anymore — it's all interaction between systems.
Nu makes interaction the primitive.
- Ref: a name for a value, wherever it lives. A KV slot, a text block in the browser, an LLM endpoint, a remote object.
- Interaction: what you do with a Ref. Read, write, branch, iterate, compose.
- Fabric: binds Refs to a real system. Bind as many as you need, they all speak the same primitive.
Same primitive, different system. One Ref for any resource, one Interaction for any op. Setting a browser text block is the same interaction as setting a KV slot.
Nu is the Python implementation of the interaction model.
Four fabrics from nustd, the same code shape in each.
Persistent state. Stores any Python type.
Reach any value by name; it survives restarts. Same code local or sharded across a cluster.
- Serverless, scales to terabytes
- Store any Python type
import nu
import nustd
class DB(nu.Shape):
hits = nustd.kv.IntRef.slot()
# +1 to a persistent counter, starting at 0
op = DB.hits.init(0) >> DB.hits.inc() >> nu.print(nu.str(DB.hits))
app = nu.With(
nustd.kv.rocksdb_navigator(".db"),
body=nustd.kv.auto_flow_atomic(op),
)
nu.run(app) # 1, then 2 on the next run, then 3Powered by virtuals · rdbpy · RocksDB · LMDB.
Reactive UIs from Python. No JS, no build step, no websocket you had to write.
A text block, a chart, a form — set them like variables, they render. Update the value, the browser updates itself.
- 50+ components out of the box
- Live updates for free
import asyncio
import nu
import nustd
class Home(nustd.ui.Page):
hello = nustd.ui.TextRef.slot()
# renders live in the browser
op = Home.hello.set("Hello, browser.")
asyncio.run(nu.arun(nustd.ui.serve(Home, op)))Same code runs local or across the cluster. No worker pool to run.
Teleport any Nu tree to any worker; it runs there and returns the result. Where it runs is a binding, not a rewrite.
- Same code, local or remote
- Distribute with a single line
import asyncio
import nu
from nustd.cluster import RayCluster, RayService, Teleport
# any Nu tree
op = nu.print("hello from a worker")
# a local ray instance with two workers; "auto" joins your cluster instead
app = nu.Provide(RayCluster, {"address": None},
nu.ProvideList(RayService, [{"num_cpus": 1}, {"num_cpus": 1}],
nu.Parallel(
Teleport(op, target=0),
Teleport(op, target=1),
),
),
)
asyncio.run(nu.arun(app))Powered by Ray. For local processes, nustd.mp and nustd.mp_pool take the same shape.
One wire, N providers. Ollama, OpenAI, OpenRouter, Groq, xAI — same call.
LLM chat as a Ref. Swap the model string, keep the code. Local models and hosted APIs meet at the same interface.
- Add intelligence to any Nu app
- 7 providers out of the box
import nu
import nustd
class Bot(nu.Service):
chat = nustd.llm.ChatRef.method(temperature=0.7)
# ask the model
op = Bot.chat(prompt="one-line haiku about rust")
app = nu.With(
nustd.llm.ollama(Bot, host="localhost", model="qwen2.5:7b"),
body=nu.print(nu.str(nu.dict(op)["text"])),
)
nu.run(app)HTTP, remote fabrics over the network, Python objects as Refs, Claude Code sessions, local worker pools, websocket servers. Same model, same shape. See all of nustd.
Python 3.10 or newer.
pip install "nustd[all]"nustd pulls in nucore, the kernel. [all] adds every backend.
Run one from examples/.
- A live counter, persistent across restarts
python examples/counter.py - An infinite series, live-sampled into a line chart
python examples/sampled.py - A movie tracker: form, filterable table, detail pages
python examples/movies.py
- Your first app: a tutorial from a counter to a live page.
- Docs: tutorials, how-tos and the full reference.
- Examples: every demo above, plus more programs to take from.
The kernel, nucore, holds the language, the engine and one fabric of its own: nu.mem, in-memory state over plain dicts. Everything that talks to the outside world lives in nustd.
Fabrics. Bind one and every Ref inside the bracket reaches a real system.
| Module | What | Extra |
|---|---|---|
nustd.kv |
Persistent state over RocksDB, LMDB or SQLite | kv |
nustd.ui |
Live web UI | ui |
nustd.cluster |
Run trees across a Ray cluster | cluster |
nustd.mp / mp_pool |
Run trees in local worker processes | mp |
nustd.llm |
OpenAI-compatible chat, seven providers | llm |
nustd.cc |
Claude Code sessions as Refs | cc |
nustd.http |
HTTP calls | http |
nustd.proxy |
Any fabric, used from another process | proxy |
nustd.service |
Python objects as Refs | - |
nustd.queue |
A queue between the event loop and threads | queue |
nustd.ws_server |
A websocket host | ws_server |
nustd.valkey |
A private Valkey server, for change notifications across processes | valkey |
The standard library. Python's stdlib, same names and call shape, except a call returns a tree that composes into the rest of the program: math, cmath, random, time, datetime, itertools, functools, asyncio, logging, pathlib, uuid, decimal, fractions, plus Nu's own fin. No extra needed.
pip install "nustd[all]" # everything
pip install "nustd[kv,ui]" # just what you needFull list in the nustd README and the reference.
Nu implements the interaction model, a language-agnostic specification of what an interaction is, how Refs name locations and how Interactions compose into programs.
Building on Nu? Add the badge to your README:
[](https://github.com/nustackdev/nu)
Open an issue or a PR. The codebase is small and readable, and the model is stable enough to build on.
Apache-2.0. See LICENSE.md.