What every agent should know

Keep your project's direction in human hands as agents do more of the work.

Join a study: we review your next ten PRs against your project’s commitments.

Your decisions should survive the next PR.

Give us one hour and read access to your repository. We’ll review your next ten PRs against your project’s commitments.

We’ll flag conflicts and consequential choices that still need your decision. You decide which findings matter. Your feedback helps us learn where the review helps.

Discuss a review

Email a repository link and a sentence about how you use coding agents. For private repositories, we’ll discuss access first.

Watch Nauro work with your agents.

A decision approved with Codex carries into Claude Code, where it stops an unsupported increase to the concurrency limit.

How Nauro works.

Retrieve prior decisions.
Your agent reads the project’s current state and the reasoning behind earlier choices.
Surface relevant commitments.
Before choosing an approach, your agent checks the proposal against related decisions.
Approve consequential choices.
You settle what should guide future work and approve what enters the record.

How the record works

Without human oversight, AI can drift.

What's the current state of my project?
What changed since yesterday?
Have we decided this before?
Why did we do it this way?
Does this conflict with something we agreed on?
Let's come back to this later.
Let's lock this in.
Here's where we got to today.

Nauro MCPget_context

Returns current state, open questions, and recent decisions.

Nauro MCPdiff_since_last_session

Returns what was recorded since the last snapshot.

Nauro MCPsearch_decisions

Finds prior decisions by relevance across the whole record.

Nauro MCPget_decision

Returns the full decision with its rationale, rejected alternatives, and supersession history.

Nauro MCPcheck_decision

Surfaces overlapping decisions before the agent picks an approach, without writing anything.

Nauro MCPflag_question

Files an open question so it is neither lost nor guessed at.

Nauro MCPpropose_decision

Records the decision after you approve the draft in chat.

Nauro MCPupdate_state

Records where things stand so the next session starts current.

Nauro gives connected AI the tools to stay consistent.

Nauro MCP get_context

Returns current state, open questions, and recent decisions.

Nauro MCP diff_since_last_session

Returns what was recorded since the last snapshot.

Nauro MCP search_decisions

Finds prior decisions by relevance across the whole record.

Nauro MCP get_decision

Returns the full decision with its rationale, rejected alternatives, and supersession history.

Nauro MCP check_decision

Surfaces overlapping decisions before the agent picks an approach, without writing anything.

Nauro MCP flag_question

Files an open question so it is neither lost nor guessed at.

Nauro MCP propose_decision

Records the decision after you approve the draft in chat.

Nauro MCP update_state

Records where things stand so the next session starts current.

Explore the record.

Browse the decisions, reasoning, and history your connected agents can retrieve.

Get started

Connect MCP Install CLI Get macOS app

Hosted MCP

https://mcp.nauro.ai/mcp

Add the URL as a custom connector and sign in.

Open connection guide

Install Nauro in your terminal.

Need uv? Install it first. It also installs Python.

1. Install the CLI

uv tool install nauro

2. Adopt this repository Run inside your Git repository

nauro adopt

Open CLI quickstart Install uv

Use the guided macOS setup.

Optional: the CLI and MCP work without it. Nauro guides you through repository and agent setup on first launch.

Runs onmacOS, Apple silicon

AccountNot required

Download for macOS View app guide

Local by default. Local use keeps your record in Markdown and does not require an account. Cloud sync stays off until you enable it.

How local storage works

How is this different?

From AGENTS.md or CLAUDE.md?

Those files carry standing instructions, not history. Nauro writes your AGENTS.md context section from the record and keeps what a flat file cannot hold: each decision’s rationale, rejected alternatives, and supersession chain, retrievable by any connected agent.

From Cursor rules or agent memory?

Rules and memory belong to one tool. The record is vendor-neutral local Markdown, read over MCP by Claude Code, Codex, Cursor, and anything else you connect. Nothing enters it without your approval.

From committed ADRs?

ADRs wait to be reread. Nauro decisions are retrieved: before a change, agents check the proposal against the record and get the related history ranked by relevance.

What happens when a decision goes stale?

It is superseded, never silently edited. The old decision keeps its number and points forward, the new one carries the reasoning, and you approve the change. Agents always see the current rule with its lineage.

When do I not need this?

If committed decisions plus an agent instruction keep one repo aligned, keep them. Nauro is for judgment that must stay current across sessions, agents, and tools.