01 / Language
Write agents, don't wire them
Declare agents, MCP servers, tools, sandboxes, and workflows in a clean, YAML-like format. No boilerplate, no glue code.
A language for agent ecosystems
Describe models, agents, MCP servers, sandboxes, tools, skills, and workflows in one clean .nt file. One command brings it all up.
age.nt — example
Explorer
⌄ example
⌄ mcp
⌄ subagents
age.nt
1# age.nt — the entry file for this ecosystem.
2# It imports every capability, then wires them onto one agent.
3
4import ./config.nt
5import ./sandboxes.nt
6import ./skills.nt
7import ./tools.nt
8import ./mcp/history.nt
9import ./subagents/researcher.nt
10
11agent age
12 description: Estimates a person's most likely age from clues.
13 model: anthropic/claude-sonnet-5
14 thinking: high
15 sandbox: workspace
16 skills:
17 - estimation
18 tools:
19 - current_year
20 - history.search_events
21 - fs_write
22 subagents:
23 - researcher
24 instructions: |
25 You estimate a person's most likely age from the clues provided.
26 Call current_year whenever a clue implies a birth year.
27 Search the trusted history server when a clue names an event.
28 Delegate to the researcher to pin down the year of an event.
29 Follow the estimation checklist, then commit to one integer.
30 input:
31 clues:
32 type: string
33 description: Any details that hint at the person's age.
34 output:
35 age:
36 type: number
37 description: The single best age estimate.
38 reason:
39 type: string
40 description: A one-sentence justification.
NTUTF-8Spaces: 2Ln 40, Col 1
Why NT
One language for the whole agent stack.
Keep the model, tools, MCP connections, delegation, and runtime behavior together in a project you can read at a glance.
02 / CLI
One CLI to run it all
Validate, graph, run and chat. A single nt command brings your whole ecosystem up against a live model.
03 / MCP
Connect MCP servers safely
Use local or remote MCP tools with fingerprinted trust, explicit selection, schema validation, and per-call approvals.
04 / Authoring
As easy as writing English
Describe what an agent should do in plain language. The engine wires the model, tools and delegation for you.
Install the CLI and run your first agent in minutes. Build your first agent →