nogra✓
Install free

{ Install · Claude Code }

Install Nogra.

The public install runs inside Claude Code — on the Claude plan you already pay. Two commands, then everything stays local.

Public marketplace source: github.com/nograai/nogra-claude-marketplace · Apache-2.0

  1. 01

    Prerequisites

    Claude Code installed — that's it. Node ships with Claude Code and powers Nogra's local runtime. No API key, no Python, no separate services — the plugin installs nothing of its own beyond what Claude Code already gives you.

  2. 02

    Add the marketplace

    Registers Nogra's public marketplace source with your Claude Code installation.

  3. 03

    Install the plugin

    Installs the public Nogra plugin into Claude Code.

  4. 04

    Verify it's installed

    Shows the plugin version, workspace id, recent local records, and checkpoint freshness. If you see a plugin ref, you're good.

  5. 05

    Initialize your workspace

    Scaffolds your workspace: .nogra/ (the local trust store — briefs, dispatch receipts, evidence, checkpoints, decisions, current tasks), plus CLAUDE.md, the brain/ vault, inbox/, and projects/. Plain markdown and JSON — you can read it, you can git commit it. Durable memory lives outside this, in Claude Code's own native Auto Memory.

{ Transparency }

What gets installed

Exactly five things land on your machine:

  • Commands and skills — the /nogra:* workflow surface inside Claude Code.
  • Hooks — the gate and the observers. Deterministic, readable source, no model judgment in the decision path.
  • The memory bound — Claude Code keeps its own native Auto Memory; Nogra's hook reads it read-only at session start and nudges you to consolidate once it grows past 200 lines or 16,000 characters. Nogra keeps no copy and scaffolds nothing here — Claude writes the facts, the hook owns the bound.
  • A local runtime — the Node script that owns briefs, receipts, and the ledger under .nogra/.
  • The brain/ vault — a pull-first knowledge base in your workspace: raw/ sources compile into wiki/ pages, indexed for deep work. Loaded only when you pull it in, never every session. /nogra:brain-init scaffolds it back if you remove it.

And what doesn't: no telemetry, no cloud calls, no credential access. Auth stays first-party — Nogra never sees your login, and nothing leaves your machine until you turn on sync.

{ First brief }

Your first brief

Type /nogra:brief followed by what you want to do. Nogra drafts a brief from your intent, you review the scope and evidence contract, then you say GO when you’re ready.

Dispatch runs a fresh executor against the brief. Verify checks the returned output against the contract. Both steps leave a receipt in .nogra/ that you can git log.

The gate ships safe: auto-approval is off until you opt in — nothing runs unattended by default, and some doors never open automatically at all.

The work leaves a record, not a memory.

{ Leaving }

The exit is part of the trust.

One command, and Nogra is gone:

Your .nogra/ files stay exactly where they are — plain markdown and JSON, readable without Nogra. The briefs, receipts, and decisions you built are yours, plugin or no plugin.

No lock-in is a feature, not a promise.

{ Troubleshooting }

  • /nogra:status— inspect installed version, workspace state, and freshness.
  • /nogra:update— refresh Nogra guidance without starting work.
  • /reload-plugins— if a command is missing after install, reload and check status.
  • MCP tools missing?— first start downloads the server via npx (one time, ~20 MB); give it a moment and retry. Hooks and commands work fine in the meantime.
  • github issues— file a bug or ask a question.