FAQ¶

Do I write the spec and the roadmap, or does N45?¶

N45 does. You describe what you want in plain language; it runs the discovery, proposes the spec (flows, screens/contracts, error cases) and generates the roadmap with tasks. You approve at each step — the intent, the spec, the roadmap — and adjust anything you want before any code is written.

Why use N45 instead of just chatting with the AI in my IDE?¶

Chatting directly works for small changes, but the AI forgets context between requests, doesn't follow your patterns consistently, and doesn't review its own work. N45 adds structure (discovery → spec → roadmap → execution), review by default (a reviewer per task), memory (STACK/PATTERNS/PROJECT live in .n45/ and feed future runs) and you in control (every relevant decision stops to ask).

What's the difference between the quick path and the structured path?¶

For a project feature, N45 decides during triage: a small, clear change (no new contract, no migration, no architectural change) → quick path (applied straight on a branch). Anything bigger → structured path (discovery + spec + roadmap + phased execution). Workspace features are always structured (they have a CONTRACT). See Build a feature.

Does N45 cost anything on top of my AI subscription?¶

The AI calls use your own IDE's connection to your provider (Anthropic, OpenAI, etc.) — N45 doesn't bill tokens separately or broker the model. What N45 requires is an active license to use the product. So: you pay for your AI (as you already do) + the N45 license.

Which AI model should I use?¶

Use the most capable and recent one available in your IDE (in the Claude family, the newest model). N45 orchestrates several subagents with strict contracts, and the quality of the result tracks the quality of the model. Weaker models work, but deliver less.

Does N45 edit my code on its own?¶

Not directly. The orchestrator never writes code — it delegates to specialized subagents that work in isolated worktrees (separate working copies). You see the result, review it and approve before the merge. Nothing reaches your main code without going through you.

Does N45 run tests?¶

Yes. Testable behavior (an endpoint, a business rule, a calculation, a validation) always gets a test — it's part of each task's acceptance criteria, and a reviewer checks it. "No tests" only applies to a genuinely throwaway prototype, and with your explicit confirmation.

What happens if a task gets stuck or fails?¶

N45 doesn't muddle through. If a task can't complete (an external dependency, ambiguity, an infra error), it's blocked and N45 escalates to you to decide — instead of shipping broken code. Problems outside the scope become recorded tech debt, not silent fixes.

Can I adjust the spec or roadmap after they're generated?¶

Yes. You review and adjust the spec before approving. During execution, if you realize something must change, N45 recomputes the affected part (a workspace spec is immutable while a roadmap is active, but change-spec re-plans only what changed).

Can I resume a roadmap later, in another session?¶

Yes. All the state lives in your repository's .n45/ (roadmap, tasks, worktrees). Close the IDE, come back weeks later, run /n45 — it reads the state from disk and continues where it left off. In a workspace, the state is read from each member's disk.

Does N45 touch my git?¶

Yes, in a disciplined and transparent way. Each feature is born on its own branch (off a protected branch, or your current branch if you choose). Subagents work in worktrees. At the finish you choose merge or Pull Request and the target protected branch. Nothing happens without your confirmation.

Will it push or open a PR without me asking?¶

No. When finishing, N45 asks whether you want a merge or a PR and which protected branch to target. The push itself stays with you. It never merges into a protected branch or opens a PR without your choice.

What is the `.n45/worktree/` folder for?¶

That's where N45 creates worktrees — isolated working copies where each task runs without disturbing your main tree. They're temporary: cleaned up after the merge (or via cleanup-worktree). The folder is gitignored.

Can I use N45 on an existing project?¶

Yes. The first time, N45 analyzes your code and generates the foundational docs (STACK, PATTERNS, PROJECT, RUNBOOK) — the "memory" it uses to respect your conventions. From then on, each feature builds on that context.

Will N45 change my patterns or my stack?¶

No. On an existing project, N45 preserves what's already in use — it doesn't migrate your language, framework or database. What the feature introduces as new follows your patterns (and, if there are organization policies, conforms to them). The rule: preserve the existing, conform the new.

When should I use a Workspace instead of a Project?¶

Use Project when the work lives in a single repository (even a monorepo — that's one repository). Use Workspace when your product lives across several repositories and you want features that span them as one thing. See What is a Workspace.

How do I update N45? And what's "run upgrade first"?¶

Run npx n45-ai@latest upgrade — it migrates the .n45/ and refreshes the skills (in a workspace, it runs in each member). If you try an init with a different version from the one that set the project up, N45 stops and asks for the upgrade first: this guarantees the .n45/ is migrated before the new skills run against an old schema.

Does N45 send my code anywhere?¶

N45 itself doesn't send your code anywhere. The AI calls go through your IDE's existing connection to your provider (Anthropic, OpenAI, etc.). N45 only sees the responses, never your code as a separate transmission. The data it keeps (.n45/) lives in your repository. It's yours.

Can I switch between Claude Code, Cursor and Codex?¶

Yes. Run npx n45-ai@latest init again in the same project from the other IDE — N45 detects the new tool and installs the matching configuration alongside the existing one. You can use all three interchangeably.

Which languages does N45's analysis support?¶

Any language. N45 reads source code as text and reasons about it using your AI assistant's model. Documents like STACK and PATTERNS reflect what the model can interpret — typically excellent for mainstream languages (Python, TypeScript, Go, Java, C#, Rust, Ruby, etc.) and reasonable for the rest.

Does it work in monorepos?¶

Yes. Run npx n45-ai@latest init at the root of the monorepo — N45 reads patterns across all packages. If your product lives across several repositories (polyrepo), use a Workspace, which coordinates the repositories as members and plans/executes features that span all of them, from a single session.

Does it work offline?¶

No. N45 needs the network to validate your license (an occasional check), make the AI calls (via your IDE) and download updates. The binary runs locally, but the workflows that drive agents require a connection.

Can I use N45 in CI?¶

Not at the moment. N45 is designed for interactive sessions inside your IDE, where the human makes the decisions. CI use cases (auto-generated PRs, scheduled refactors) are under evaluation but not yet supported.

What happens when I uninstall N45?¶

There's no formal uninstall — just delete .n45/ from your project. The IDE config files (CLAUDE.md, .cursor/rules/, AGENTS.md) keep the N45 block between  and  markers. Remove that block (or the whole file if N45 created it) and you're done.

My IDE isn't supported. What can I do?¶

Today N45 supports Claude Code, Cursor and OpenAI Codex. If you'd like support for another tool, write to support@n45.tech with your use case.