naetive

Starting a chat that already knows the project

Last reviewed: 16 Sep 2026

You never paste a long prompt. An agent connected to Naetive can orient itself in one call. This page is the short version of what to say, and what to put in a repository so Claude Code does it without being asked.

Three ways a chat starts

  • New project: "Set this up in Naetive: one line about it." See Start a project with your AI.
  • Existing project: "Get oriented on project first, then …" A set-up agent orients itself: it calls get_project_context for the brief picture, which folds in the understanding digest (identity, purpose, current focus, key decisions, constraints) and recent activity, then narrows to your task. It inherits whatever earlier sessions established.
  • Normal chat: do not mention Naetive and it stays a normal chat.

For Claude Code: two files

Global, ~/.claude/CLAUDE.md: the one-time instruction from Start a project with your AI. It makes Claude Code Naetive-aware everywhere without assuming every folder is a Naetive project.

Per repository, ./CLAUDE.md in a repo that is a Naetive project: an orient line, because here Naetive is the project's memory.

This project is tracked in Naetive (slug: your-slug). At session start, call
get_project_context({ project: "your-slug", format: "brief" }), then narrow
with a scoped get_project_context for the task. After a context compaction,
call get_session_recovery_context() first.

If the repository carries the .naetive folder, the export already wrote a pointer to it in AGENTS.md and CLAUDE.md, so an agent with no connection at all still reads the understanding first.

For everyone else

ChatGPT, Claude on the web and Claude Desktop take the same one-time instruction in their custom instructions. Codex reads AGENTS.md. Then "get oriented on project first" is all you need to say.

New = "Set this up in Naetive." · Existing = "Get oriented first." · Otherwise = just chat.