henergyqueai

The HenergyqueAI coding assistant in your terminal. It reads your project, suggests changes and runs commands, always within the permission mode you choose. Requires Node.js 22 or later.

The assistant answers in your language. Its commands, options and settings use French names (/aide, --modele, demander…): this page lists them all with their meaning.

Install

npm install -g henergyqueai

Sign in

Nothing to set up. Just run it:

henergyqueai

Your browser opens, you click Autoriser (Authorize), and you're done. No key to copy or paste. You need a free account on henergyqueai.fr.

The first time you run it in a folder, it asks whether you trust that folder. Answer, and the screen clears to make room for the session.

During a session, the interface fills the terminal: the butterfly stays at the top, the answer builds up in the middle and your input stays visible at the bottom. Press Enter on an empty line, ↑ in an empty field or Ctrl+K to open the actions menu. Type in the menu to filter, then use ↑↓, Enter and Esc to pick or go back. Shift+Enter adds a new line to your message; Page Up/Down scrolls through the history. While an answer is streaming, Esc stops it.

The Joindre un fichier (Attach a file) menu inserts a reference. A reference such as @src/app.js (or @{project notes.md} for a name with spaces) attaches an excerpt of the file to your request, up to five files of 20,000 characters each. A reference to an image (@mockup.png) sends it as an image. References pointing outside the project are ignored.

| Shortcut | What it does | | ------------------ | ------------------------------------------------------------------- | | Esc Esc | Go back to an earlier request (conversation, code, or both) | | Ctrl+O | Detailed view: full tool results | | Ctrl+R | Search your request history | | Alt+V / Ctrl+V | Paste a screenshot from the clipboard | | Shift+Tab | Switch mode (ask, accept edits, plan) | | # instruction | Adds the instruction to HENERGYQUEAI.md for future sessions | | !command | Runs a command in the project's terminal |

An answer that takes more than 30 seconds rings the terminal bell when it finishes or needs your approval. When less than 30% of the context is left, the footer says so: /compact frees it up.

Use

henergyqueai                    start a session
henergyqueai "fix the date bug in invoice.js"
cat error.log | henergyqueai "where does this error come from?"
henergyqueai -c                 resume this folder's conversation

It only has access to the folder you started it from.

The assistant can find files by pattern (src/**/*.js), read and search text, propose a targeted edit with a preview, show the Git status or diff, and run your tests. Writes and commands follow the mode and permission rules you chose. To give it project-specific instructions, create HENERGYQUEAI.md, AGENTS.md or .henergyqueai/instructions.md at the project root.

It can also search the web and read a page (20 searches a day on the free plan, 80 with a subscription), ask you a multiple-choice question when a request is ambiguous, run a long command in the background (dev server, tests in watch mode) and read its output, and hand independent research to read-only sub-agents that work in parallel. Each sub-agent step counts as one request. In plan mode, it presents its finished plan: you accept it (with or without approving each change) or send it back to work.

It can track long tasks in a persistent, per-project list. The Tâches (Tasks) menu lets you review and complete steps; /tache ajouter <title> adds one. Local skills can live in .henergyqueai/skills/<name>/SKILL.md or ~/.henergyqueai/skills/<name>/SKILL.md. The assistant can read them and they appear in the Compétences (Skills) menu; scripts are never run automatically.

Local MCP servers can be declared in ~/.henergyqueai/mcp.json. The CLI connects them at startup and asks for approval before each tool call, unless the mode or a rule already allows it. Example:

{
  "servers": {
    "my-tool": {
      "command": "node",
      "args": ["/path/to/mcp-server.js"],
      "env": { "ACCESS_TOKEN": "${MY_LOCAL_TOKEN}" }
    }
  }
}

The token stays in your local environment variable. The Outils MCP (MCP tools) menu shows connections and their errors. A rule such as Mcp(my_tool_*) can allow a trusted server's tools; a deny rule always wins.

Options

| Option | What it does | | -------------------------------- | -------------------------------------- | | --modele <aelyr\|oryne\|velys> | Picks the assistant | | --mode <mode> | See the modes below | | --continuer, -c | Resumes this folder's conversation | | --direct, -p | Answers without saving anything | | --lecture-seule | Shortcut for --mode lecture-seule | | --oui | Shortcut for --mode sans-demander |

The three assistants: Velys for everyday questions, Oryne for code, Aelyr, the most capable, for demanding projects (Envol and Horizon plans).

Modes

| Mode | What it may do | | ------------------------ | --------------------------------------------------- | | demander | Asks before writing or running anything. Default. | | accepter-modifications | Writes without asking, asks before commands | | lecture-seule | Read-only: can neither write nor run | | plan | Explores and proposes a plan without changing files | | sans-demander | Never asks |

In a session

The keyboard menu is the main way in. The commands below remain available as shortcuts and for non-interactive terminals.

| Command | What it does | | ------------------------------- | ------------------------------------------------ | | /aide | Opens the menu (help) | | /modele <aelyr\|oryne\|velys> | Switches assistant | | /mode <mode> | Changes what it may do | | /statut | What is loaded: settings, rules, exchanges | | /contexte | Estimated context usage and instructions | | /utilisation | Plan and requests used today | | /doctor | Local diagnostics and service connection | | /theme | Picks the terminal colors | | /taches | Opens the task list | | /tache ajouter <title> | Adds a task | | /tache finir <id> | Completes a task | | /competences | Lists installed skills | | /competence <name> | Shows a skill | | /mcp | Shows connected external tools | | /git | Git repository status | | /diff | Git changes | | /revue, /review | Reviews your changes without modifying anything | | /compact | Summarizes the history to free up context | | /fichiers | Lists the project's files | | /sessions | Resumes a saved conversation | | /effacer | Forgets the conversation | | /plan | Enters or leaves plan mode | | /init | Analyzes the project and prepares HENERGYQUEAI.md | | /rembobiner, /rewind | Goes back to an earlier request | | /exporter [file] | Copies the conversation as Markdown, or writes it | | /processus | Background commands: output and stop | | /hooks | Shows the loaded hooks | | /quitter | Quits |

To switch accounts, quit the session and run henergyqueai connexion.

Settings

Two files; the closest one wins:

| File | Scope | | -------------------------------------- | ------------------ | | ~/.henergyqueai/settings.json | You, everywhere | | <project>/.henergyqueai/settings.json | This project |

{
  "modele": "oryne",
  "mode": "demander",
  "permissions": {
    "autoriser": ["Executer(npm run test *)", "Lire"],
    "refuser": ["Executer(rm *)", "Ecrire(.env*)"]
  }
}

autoriser lists allowed actions and refuser denied ones. A rule names a tool — Lire (read), Ecrire (write and targeted edit), Executer (run), Lister (list and find files), Chercher (search), Git or Mcp — with an optional pattern where * matches anything. A deny rule always wins over an allow rule.

An allowed rule skips the question. A denied rule stops the action before it happens.

Hooks

Your own commands, run around the assistant's work, in the same settings files:

{
  "hooks": {
    "avant_outil": [{ "outils": "executer", "commande": "node check-command.js" }],
    "apres_outil": [{ "outils": "ecrire_fichier|editer_fichier", "commande": "npx prettier --write \"$HENERGYQUEAI_FICHIER\"" }],
    "fin": [{ "commande": "npm test --silent" }]
  }
}

avant_outil runs before a tool, apres_outil after it, and fin when the assistant finishes. outils is a regular expression on the tool name and commande is the command to run. Each command receives HENERGYQUEAI_EVENEMENT (event), HENERGYQUEAI_OUTIL (tool), HENERGYQUEAI_FICHIER (file), HENERGYQUEAI_COMMANDE (command) and the full details as JSON on its standard input. Exit code 2 blocks the tool (avant_outil), passes its error output back to the assistant (apres_outil) or sends it back to work (fin, at most twice). A project's hooks only run in a folder you have trusted.

Going back restores the files the assistant changed, not the effects of commands run in the terminal.

Environment variables

| Variable | Purpose | | ------------------ | --------------------------------------------------------------------------- | | HENERGYQUEAI_KEY | Device sign-in token; takes priority over the one from henergyqueai connexion | | HENERGYQUEAI_URL | Service address, for a private deployment | | NO_COLOR | Turns colors off |

Your access

Each authorized terminal appears under your machine's name in Paramètres → Appareils connectés (Settings → Connected devices) on the website, with its last-used date. You can sign it out at any time without affecting your other machines. API keys from the developer platform are for your own applications, not for the terminal.

Also available

• VS Code: the HenergyqueAI extension, code --install-extension henergyqueai.henergyqueai.
• Your own apps: the official libraries, npm install henergyqueai-sdk and pip install henergyqueai.

License

Proprietary. © HenergyqueAI.