THE OPEN AGENT TOOLKIT

Build agents that
do the work.

The runtime, coding agent, and observability tools to take an agent from a promising demo to a dependable part of your stack.

Start building Explore the toolkit
Open source TypeScript first Your models, your rules
n.AGENT LOOP
YOUR CODE
ANY MODEL
YOUR TOOLS
LIVE EVENTS
A runtime you can inspect
ONE OPEN TOOLKIT
not-another-harness+ nah agent+ nah studio+ nah cloud soon
01 / THE FRAMEWORK

Know what your
agent is doing.

Build on an explicit agent loop with streamed events, real tool execution, and budgets you can set. Bring your own model and shape every decision around it.

Typed events from every step Spend and context budgets Tools with boundaries you control
Explore the framework
RUN TRACE LIVE
Model responsestreaming · 248 tokens
Tool executionsearchCode · 34ms
Budget check$0.014 of $1.00
Next step ready
run_01J8…3 events · 1 tool
02 / THE CODING AGENT

Put the harness
to work.

nah is a terminal coding agent built on the same framework. Work in your repo, keep the loop visible, and decide what the agent can change.

Interactive or headless Permission modes you choose Sessions that stay with your project
Meet the nah agent
agent.ts RUNNING
01import { createAgent } from "not-another-harness"
02import { openai } from "@ai-sdk/openai"
03
04const agent = createAgent({
05 model: openai("gpt-4.1"),
06 tools: [searchCode, editFile],
07 maxSteps: 12,
08});
09
10for await (const event of agent.run(task)) {
11 console.log(event.type);
12}
events streamingTypeScript · 12 lines
LIVE RUN00:02.48
run.startedtask accepted
tool.calledsearchCode
model.streamingthinking…
FROM FIRST RUN TO PRODUCTION

One loop. Every place
you need it.

Develop locally, understand every run, then bring the same agent into your application.

See how it works
01 / BUILD

Make it yours.

Compose the runtime with your model, tools, and policies. Keep the parts you need.

Read the quickstart
02 / OBSERVE

See the whole run.

NAH Studio brings traces, spend, evaluations, and agent conversations into one local dashboard.

Explore Studio
03 / SHIP

Take it anywhere.

Use the same runtime in a CLI or mount it behind a web request handler. No platform lock-in.

Explore the SDK
NAH STUDIO / INCLUDED

Observe the work.
Keep the control.

A local dashboard for agent runs, traces, spend, and evaluations. Studio is opt-in: when it isn’t running, your agent doesn’t phone home.

$ npx --yes nah-studio ↵
Explore NAH Studio
NAH STUDIO LOCAL
Recent runsAll agents⌄
AGENTSTATUSSPEND
code-reviewercomplete$0.018
nahcomplete$0.042
docs-writerrunning$0.009
TOTAL SPEND $0.069
Across 3 runs today
YOUR AGENTS, CONNECTED
ON THE HORIZON

NAH Cloud

For teams ready to run agents together. Centralize traces, manage shared environments, and give every teammate a clear view of what’s happening.

We’re building the hosted layer now. The open runtime and local Studio work today.

Get the NAH Cloud preview
YOUR AGENT, YOUR CALL

Start with a loop
you can see.

Build your own agent, try nah in a repo, or open the docs and explore.

Start building Read the docs