nt

README

nt

A terminal task & note manager that stores everything as plain files — todo.txt tasks + markdown notes — so your editor, grep, git, and AI coding agents can all read and write it directly. Built to be the durable memory layer for AI sessions: action items an agent creates survive the session in text the next agent can read back.

See SPEC.md for the full design.

Status: all four phases implemented — core + CLI + AI loop (P1), the Bubble Tea TUI (P2), recurrence + dependencies (P3), and the Claude Code hook + /nt skill (P4).

Install

# Curl — downloads the latest release binary to ~/.local/bin (no Go needed)
curl -fsSL https://raw.githubusercontent.com/navbytes/nt/main/install.sh | bash

# Go (installs the latest tagged release to $GOBIN)
go install github.com/navbytes/nt@latest

# From source — builds and installs to ~/.local/bin (override with NT_INSTALL_DIR)
git clone https://github.com/navbytes/nt && cd nt && make install

Releases are automated: pushing a vX.Y.Z tag runs GoReleaser via GitHub Actions to build the cross-platform binaries + checksums (RELEASING.md). Homebrew is planned for later (brew install navbytes/tap/nt).

Pure Go, single static binary, no system dependencies. nt --version reports the build.

Build (development)

go build -o nt .       # Go 1.24+
./nt                   # launch the TUI
./nt help              # CLI reference
make test              # run the test suite

TUI

Run nt with no arguments for the interactive terminal UI (Bubble Tea). It adapts to terminal width — compact strip / standard list / wide split with a live detail pane — and live-refreshes via fsnotify when another process (a CLI call, an AI session) writes the store. Three tabs: tasks, notes, and a Logbook of completed work grouped by completion date. Mouse clicks select rows, activate [[link]]/@tag/+project tokens, and switch tabs. Press ? for the full keymap. Essentials: j/k move · enter detail · x done · a/A add · space/V mark · X delete · p priority · D due · t tag · l/L link/follow · y yank · / filter · v group · . show done · 1/2/3 tasks/notes/logbook · Ctrl+l lock (read-only) · u undo · q quit.

Screenshots

Tasks — wide split	Logbook — completed work by date
Tasks — done hidden (✓ N done chip)	Notes

More views in docs/screenshots/ — regenerate with ./scripts/screenshots.sh.

Store

One global store at $NT_DIR (default ~/.local/share/nt):

~/.local/share/nt/
├── tasks.txt     # todo.txt format, one line per task
├── done.txt      # archived completed tasks
├── undo.jsonl    # undo transaction journal
└── notes/*.md    # markdown notes with YAML frontmatter

Everything is plain text — open it in any editor, grep it, or git init it. To version-control the store, run nt git-init (sets up merge=union so branches don't conflict on every add, plus a .gitignore for transient files); after a merge, nt doctor reconciles any duplicates.

Usage

nt add "fix auth bug" --pri high --due today --tag backend --project api
nt note "JWT expiry" --body "tokens last 24h" --tag auth
nt note "Auth design" --folder work          # file it in notes/work/ (or: nt note "work/Auth design")
nt ready [--source claude] [--json]   # open, unblocked tasks by urgency — start here
nt list [--status open|doing|blocked|done] [--tag T] [--sort urgency] [--all] [--json]
nt done <id|task:N>            # also accepts the 6-char short code shown in list
nt update <id|task:N> --status doing --due tomorrow --pri med
nt add "weekly review" --due monday --recur weekly   # completing spawns the next
nt add "write migration" --blocks task:5             # task:5 hides until this is done
nt list --show-blocked                               # reveal dependency-blocked tasks
nt search "auth"               # ripgrep over notes + substring over tasks
nt links <id|task:N>           # forward links + backlinks (both directions)
nt recall --source claude --json   # read items back — the AI loop
nt log [--since|--days N] [--json]  # completed tasks, newest first (the Logbook)
nt mcp                              # stdio MCP server — typed tools for agents
nt mcp install [--client claude-desktop]   # register nt with an AI client (absolute path, idempotent)
nt web [--port N]                   # browse & read notes in a browser (localhost, mermaid, live-reload)
nt edit <id|task:N>            # safe $EDITOR round-trip (never touches the shared file directly)
nt mv <note> <new-name|folder/path>   # rename/move a note, rewriting all [[links]] to it
nt archive                     # move done tasks to done.txt
nt undo                        # revert the last change (and undo-again to redo)
nt path                        # print $NT_DIR

Task format (todo.txt + conventions)

(A) fix auth bug +api @backend due:2026-06-05 [[jwt-expiry]] src:claude id:01JZ8RT3

• (A)/(B)/(C) priority · +project · @tag · due: · src: origin · id: ULID
• [[target]] links to any note or task; parent:/blocks: are typed task links
• Completing a task preserves its priority as pri:A; reopening restores (A)

Linking

[[ ]] cross-links tasks and notes in any direction. Backlinks ("linked from") are computed on demand by scanning files — no index to maintain.

Notes ↔ Obsidian

The trend is "use Obsidian as your AI knowledge base." The substance behind it is plain markdown — not the app. What actually helps an LLM is the Karpathy "LLM wiki" pattern: a library of plain .md the model reads and maintains directly. For a personal knowledge base that beats RAG on reliability (the model reads the real file, not an embedding's best guess), token cost (open only what's relevant), and auditability (git diff it, roll it back). nt is that pattern, with tasks and an AI-memory loop on top.

So nt talks to Claude by letting it read scoped files directly — nt recall, nt ready, and the nt mcp typed tools hand back exactly the relevant items. It deliberately does not route the agent through Obsidian's REST-API MCP, which practitioners find slow, token-heavy, and a whole-vault privacy exposure that experienced users won't run in their main vault (forum consensus). Obsidian stays an optional human GUI pointed at notes/ — never a dependency and never in the agent's path.

nt has no notes GUI of its own — and doesn't need one. Notes are plain .md + YAML frontmatter + [[wikilinks]], so you can point an Obsidian vault at the notes/ folder and use Obsidian as the GUI while nt owns tasks, the CLI/TUI, and the AI-memory loop. Folders work both ways: create into one with nt note "…" --folder work (or the path-style nt note "work/…"), move with nt mv, and nt reads back what Obsidian writes: nested subfolders, block-list tags:/aliases:, notes without an H1 (title falls back to the filename), and link variants ([[folder/note]], [[note#heading]], [[note|alias]]) resolved by shortest path-suffix — a bare name colliding across folders is flagged ambiguous rather than guessed. Rename/move is nt-native: nt mv <note> <new> (or r in the TUI notes tab) renames the file and rewrites every [[link]] to it across tasks and notes, so you don't depend on Obsidian to keep links intact. (Logseq's outliner model is not a target.)

Web viewer (nt web)

nt web starts a localhost server to browse and read your notes in a browser — a folder tree, rendered Markdown with [[wikilink]] navigation and backlinks, full-text search, Mermaid diagrams, and light/dark themes that match the TUI. It also surfaces nt's task↔note graph — tasks that reference a note, backlinks with context snippets, an in-note table of contents, and src:claude provenance — over the same notes/ files everything else uses.

Light — TOC, breadcrumb, tasks & backlinks	Dark — Tokyo Night

nt web              # pick a free port, print the URL (127.0.0.1)
nt web --port 8080  # fixed port

It's read-only and binds to 127.0.0.1 only — your notes are never exposed on the network. The page live-reloads when a file changes (via fsnotify + SSE), so editing in your editor, Obsidian, or Zettlr updates the view instantly. Mermaid is the only vendored asset (embedded gzipped); no external requests, no CDN — it works fully offline.

Claude Code integration

nt is built to be the durable memory layer for AI sessions. Two integration points:

• PostToolUse hook — nt hook mirrors Claude's TodoWrite list into the store (idempotent, src:claude, status-mapped). Wire it in ~/.claude/settings.json.
• /nt skill — teaches Claude to capture tasks/notes and nt recall prior context.

Setup and walkthrough: docs/claude-integration.md.

# during a session (hook does this automatically, or call it directly):
nt add "fix token refresh race" --source claude --tag auth
# next session — read it back:
nt recall --source claude --json

What's guaranteed (the hard parts)

• Lossless round-trip: an unmodified tasks.txt line is re-emitted byte-for-byte, preserving unknown key:value tokens from other todo.txt tools (enforced by test).
• No lost updates: every write goes through one ULID-keyed mutation engine that locks, re-reads, mutates, and atomically renames — so a concurrent nt add from an AI session is never clobbered (concurrency test included).
• Transactional undo: each change journals before-images keyed by ULID; nt undo reverses them and supports undo-again-to-redo.

Tests

go test ./...

License

MIT

Documentation

Overview

Command nt is a terminal task & note manager that stores data as plain files (todo.txt tasks + markdown notes) so editors, grep, git, and AI agents can read and write it directly. See SPEC.md for the design.