Nauro
Decision system for software built with AI
Keep your project's direction in human hands as agents do more of the work.
Nauro keeps a living project record. It combines project scope, current state, and open questions with human-approved project judgment: intent, goals, decisions, rationale, tradeoffs, and rejected paths. Project judgment is the human-ratified part of the record; context is the relevant slice of the record an agent receives for the work in front of it. Works with Claude, Perplexity, Cursor, Codex, and any MCP client.
How the loop works
- Nauro orients the agent with project scope, current state, open questions, and relevant prior judgment.
- You and the agent clarify missing intent, constraints, or tradeoffs.
- If the work needs new or revised judgment, the agent drafts it and waits for your explicit approval.
- The agent plans, recommends, or implements with that context in view.
- The agent explains how the context shaped the result, and you accept, correct, except, reopen, or supersede it in conversation.
- The agent reports meaningful completed progress as current state, so later connected agents inherit the updated state and approved judgment.
The markdown store, context summaries, BM25 retrieval, advisory checks, and optional sync support this loop. They do not replace your judgment or silently change project truth.
Install
uv tool install nauro
No uv? Install it with curl -LsSf https://astral.sh/uv/install.sh | sh (macOS/Linux) or the PowerShell line on Windows. Already on Python 3.10+? pipx install nauro (or pip install nauro) works too.
Try the demo (optional)
Pennykeep is the fictional budgeting app bundled with Nauro.
- Request: Replace envelope budgeting with spending charts to simplify onboarding.
- Prior decision: In the fictional tester history, passive tracking did not help users control spending. The team accepted more setup effort so users assign income before spending.
- Effect on the plan: An agent could suggest simpler envelope setup or ask whether to revisit the budgeting model.
Try this example without an account or agent setup. On macOS or Linux:
mkdir -p /tmp/nauro-demo && cd /tmp/nauro-demo
nauro init --demo
nauro check-decision "Replace envelope budgeting with a passive spending tracker and charts"
Look for Envelope budgeting method in the related decisions. The command retrieves the prior reasoning; it does not run an agent or change the recorded decision.
See the demo guide for PowerShell commands and more ways to explore the sample.
If a small repo plus a reliable AGENTS.md or CLAUDE.md keeps agents oriented, Nauro may be more than you need. Nauro is designed for context that must persist across longer histories, sessions, tools, repos, machines, or repeated handoffs.
nauro graph renders the store to one self-contained HTML file and opens it: a node-link map of every decision as the default view, plus drawn supersession lineage, a timeline, and a category browser. The demo store's consolidation, three retired decisions converging on the one that replaced them, draws as a fan. By default the file carries the full decision store, including each decision's body rendered as structured detail in the side panel, and lands in the store directory rather than your repo; --no-include-bodies produces a redacted titles-and-metadata artifact for wider sharing.
nauro doctor checks the store for structural defects: unparseable decision files, dangling or cyclic supersession refs, and status contradictions. It is deterministic and report-only — it never edits the store and always exits 0. It also names one repairable defect separately: a supersession recorded on the newer decision but never written back to the older one.
nauro repair is the only command that acts on what doctor names, and it acts on that one shape. When a single decision in the store claims to supersede another that is still active with no reference back, it shows you both decisions with their versions and dates, states the exact field change, and asks. Anything less clear-cut — two decisions claiming the same predecessor, a cycle, a file that will not parse — is reported with guidance and left alone. Nothing is written without your answer, and there is no flag to skip the question.
For real-project setup (nauro init / nauro adopt), cross-surface access, MCP tool reference, and architecture details, see the main project README. Don't run nauro setup from /tmp/nauro-demo; that would wire the throwaway demo into your MCP client.
For cross-surface onboarding, run nauro adopt --with-skills --with-subagents. Plain adopt installs nauro-adopt. --with-skills adds nauro-ship-task, nauro-context, nauro-loop, and nauro-interview. --with-subagents adds Nauro's planner, executor, reviewer, and tech-lead agents for Claude Code, Cursor, and Codex. Cursor stores its native project agents under .cursor/agents/.
Cursor runs nauro-ship-task natively. nauro-loop Program Delivery stays on hold.
Restart, then seed the store with /nauro-adopt in Claude Code, $nauro-adopt in Codex, or @nauro-adopt in Cursor Agent chat. Cursor requires a working durable absolute Nauro executable for user-global ~/.cursor/mcp.json; otherwise installation fails. Claude/Codex retain their existing fallback behavior. Follow the migration steps below before changing legacy wiring. Commit .nauro/config.json, .cursor/rules/nauro-*.mdc, and .cursor/agents/nauro-*.md. Keep the global MCP configuration machine-local.
Cursor Cloud Agents need separate MCP configuration at cursor.com/agents. If you use Nauro's hosted connector there, link and sync the project first.
Cursor migration
First, from the affected repository root, run nauro setup cursor. Do not change legacy .cursor/mcp.json or its tracking unless setup reports global configuration was written or already correct. If setup fails or is uncertain, keep the legacy file and entry unchanged. This verifies configuration, not a live Cursor connection.
Before either remedy below, confirm the legacy entry has the installer's owned stdio shape: only type, command, and args fields; type absent or stdio; args exactly ["serve", "--stdio"]; and command either bare nauro/nauro.exe or an absolute path whose filename is nauro/nauro.exe (case-insensitive). A key named nauro alone does not prove ownership. Preserve custom, hosted/HTTP, additional-field, and uncertain entries. Do not adapt an unrecognized entry to make it pass this check.
Only after global configuration succeeds and ownership is confirmed:
- Entire file intentionally machine-local: from the affected repository root, run
git rm --cached .cursor/mcp.json. This removes the file from Git's index and keeps the working copy. Commit that tracking change, then rerunnauro setup cursorfor guarded migration. Do not use this path for shared configuration or unrelated content that should remain tracked. - Shared tracked file: Keep the file tracked. Manually remove only the confirmed installer-owned
mcpServers.nauroentry. Preserve all other entries and configuration. Commit that targeted edit, then rerunnauro setup cursor. Do not remove the entiremcpServersobject or shared file. - Untracked owned entry: let setup perform the guarded migration. No manual untracking is needed.
Restart Cursor and verify its MCP connection after setup succeeds.
Re-running onboarding refreshes Nauro-owned workflow files and saves differing copies as backups. It leaves third-party skills and agents untouched. Pass --force-overwrite only when you do not want backups.
Why Nauro?
Nauro supports a human-ratified project-judgment loop. It captures what you decided and what you ruled out, with the reasoning, then brings related judgment into agent work. Keyword search over the decision store is one mechanism for putting prior reasoning in front of an agent at proposal time.
No model judges your decisions. The check uses deterministic keyword retrieval (BM25), is advisory, and never blocks a change. Agents draft additions, updates, and supersessions; you explicitly approve each one before propose_decision commits it in one call.
check_decision returns the related prior decisions (the related_decisions list in the response) so the agent can weigh them before proposing; Nauro ranks by keyword relevance and does not judge the proposal. On the approved propose_decision call, near-matches surface as advisory similar_decisions, and a clean proposal commits in one call. What you approve in one tool, every connected agent inherits; for example, a decision recorded in Claude Code is available later in Perplexity. The store is plain markdown in a folder you own. Run it fully locally with no account; cloud sync is opt-in.
Hosted allowance
Nauro includes unlimited local usage, unlimited projects, and 5,000 remote MCP calls per month. For higher hosted limits, contact thomas@nauro.ai. See nauro.ai/pricing for current details.
Apache 2.0 license. Part of the nauro-ai/nauro monorepo.
Named for Peter Naur, whose 1985 paper Programming as Theory Building argued the real program is the theory in the programmer's mind, not the code. Every fresh agent session is the equivalent of losing that programmer.
Metadata
Release files for nauro 1.21.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| nauro-1.21.0.tar.gz | 474.6 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| nauro-1.21.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 1.1 MB
Release files / nauro-1.21.0.tar.gz
| Download URL | nauro-1.21.0.tar.gz |
|---|---|
| Size | 474.6 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
bb0af79d8c2f96f649c17d1d43d96e60aaee73cd67b8bd28f40d9d9755e5a7f9
|
|
BLAKE2b-256 checksum How to use checksums |
5dafc418600b8a75bc196e4cbcc12643a3cdfcdb86f299e89654e17abf1a50a9
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 4, 2026.
Transparency logRelease files / nauro-1.21.0-py3-none-any.whl
| Download URL | nauro-1.21.0-py3-none-any.whl |
|---|---|
| Size | 601.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
72c715663a186fde6ae1cc7ecb05a9678e2bcedcc4675399cc15640cf848fec6
|
|
BLAKE2b-256 checksum How to use checksums |
c25faf7c2f578f74e5bbfd537d7c003c088df8759116f5795e619e41e86a57cb
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 4, 2026.
Transparency log