naetive

Start a project with your AI

Last reviewed: 16 Sep 2026

Naetive is easiest when your agent sets the project up for you. Connect an agent, give it one standing instruction, then ask it to start. From then on every chat opens already knowing the project.

The one-time instruction

Add this once to your agent's standing instructions. It is deliberately quiet: the agent only reaches for Naetive when you ask. The Connect page shows you the same text, under Make every new chat Naetive-aware, in each agent's row.

When I explicitly ask you to use Naetive, call get_bootstrap first and follow the instructions it returns. That covers "set this up in Naetive", "start a new project on Naetive", "add this to my project on Naetive", "load Naetive context", and any time I name one of my projects on Naetive.

If I talk about a project but do not mention Naetive, do not assume I want to use it. Ask me before you call any Naetive tool or write anything.

Where it goes:

  • ChatGPT: Settings, Personalization, Custom Instructions (or a ChatGPT Project's instructions, scoped to that project).
  • Claude Code: ~/.claude/CLAUDE.md, so it applies in every folder. In a repository that is a project on Naetive, also add the orient line from Starting a chat that already knows the project.
  • Codex: AGENTS.md at the project root, or ~/.codex/AGENTS.md.
  • Claude on the web and Claude Desktop: paste it at the start of a chat, or into a Project's instructions.

Two things to note. "My projects on Naetive" means any project of yours that lives here, whatever it is called; there is nothing special about the word naetive. And get_bootstrap fetches Naetive's current rules for the agent, so those rules can change without you touching this text again.

Starting something new

Say: "Set this up in Naetive: a two-line description of the idea."

The agent checks your existing projects, creates a new one (it never writes into an existing project without asking), writes the first understanding from what you told it, and creates a first job or two so there is something to start on. One sentence in, a project out.

Bringing existing code

Clone the repository to your machine, open your agent in that folder, and say: "Map this project in Naetive."

The agent reads the code, writes the orientation (what it is, why it exists, how it is built), records the decisions and constraints it can see with the files they rest on, and creates jobs from what is unfinished. Claude Code and Codex read the folder directly; a chat agent needs the tunnel running in that folder first.

What gets created

  • The project, with its own understanding, work and documents.
  • The orientation: identity, purpose, architecture, story, written by the agent.
  • The first entries: decisions, constraints, learnings, each with the file it cites where there is one.
  • Jobs, each with a spec, so the next session has somewhere to start.
  • The folder, once you ask for it: run naetive export in the repo, or download it from the Understanding page.

An example, with Claude Code

You: Set this up in Naetive: a small CLI that renames photos by the date they were taken.

Agent: I have created the project "photo-rename" on Naetive. I recorded what you told me as the identity and purpose, wrote two constraints (never overwrite a file, never touch files outside the folder you name), and created two jobs: "Read the date from EXIF" and "Rename safely with a dry run". Want me to start on the first?

You: Yes.

Every later chat, on any agent you connect, opens knowing all of that.

How the agent should behave

  • It asks before creating a project or writing into one.
  • It records what it learns as it works, not at the end.
  • It cites a file for anything about the code, and says so when it is guessing.
  • It leaves open questions as open questions, so the next agent sees them.

If an agent is not doing these, ask it to call get_bootstrap again; the rules are there.