Aderra Agent (terminal and browser)

Aderra Agent is Aderra’s own agent. It works inside the folder you start it in: it reads and edits files, runs commands, searches the web and fetches pages, using your Aderra credits. Your sessions are saved on your computer, not on our servers.

Note

Like the rest of the terminal, Aderra Agent is included from the Plus plan. On the Connect page it is listed as Aderra Harness.

Install and log in

You need Node.js 22.19 or newer, on macOS or Linux (on Windows, use WSL).

1. Install the aderra command:

   curl -fsSL https://aderra.ai/cli | sh

2. In the Aderra app, open Connect (the terminal icon in the left bar). Under Sign in, click Generate key and copy it. It starts with ak_ and is shown once.

3. Log in, and paste the key when asked. It stays hidden as you paste. Press Enter at the Server question to keep the default.

   aderra login

4. Go to your project folder and start the agent:

   cd ~/my-project
   aderra

The first start downloads Aderra Agent (about 130 MB, and again for each new version), for macOS or Linux on Intel or ARM. Each download is checked against Aderra’s signature before it is installed. After that it starts immediately and updates itself when a new version is published. Already have aderra? Run aderra update first.

Three ways to run it

Command	What it does
aderra	Aderra Agent in this terminal
aderra web	The same agent in your browser (opens http://127.0.0.1:3095)
aderra run "<task>"	Does one task, prints the answer, and exits — for scripts

Options:

Option	Works with	What it does
--model <id>	all three	Use another model for this launch, for example --model glm-5.3-flash-uncensored for the unfiltered model. The default is glm-5.3-flash. Inside the agent: /model
--yolo	all three	Full access, never ask for approval. See permission modes
--session-id <id>	aderra	Continue an earlier session. When you quit, the agent prints the command to resume it; inside, /resume lists them
--port <n>	aderra web	Serve on another port than 3095
--no-open	aderra web	Don’t open the browser; go to the address yourself

/model offers GLM-5.3 Flash and GLM-5.3 Flash Unfiltered.

Permission modes

By default the agent can edit files inside the folder you started it in, and asks you before anything outside it. When it asks, choose Allow once or Reject.

Mode	Files it can change	Asks before
read-only	none	anything that writes
workspace-write (default)	this folder	anything outside it
danger-full-access	anything your user account can	nothing

Switch any time inside the agent with /permission <mode>. /permission on its own shows the current mode.

YOLO mode

aderra --yolo starts in danger-full-access: no sandbox and no approval prompts. The agent can delete or overwrite anything your account can reach. Use it in a throwaway folder, a container or a VM, and switch back with /permission workspace-write.

Commands inside the agent

Type / to see them with autocomplete.

Command	What it does
/help	Keyboard shortcuts and every command
/model [model]	Show or switch the model for this session
/permission [mode]	Show or switch the permission mode
/plan [off]	Marks the session as planning. It does not stop the agent from editing files: to stop edits, use /permission read-only
/goal <objective>	Give it a long-running goal to keep working toward (/goal pause, resume, clear)
/compact	Summarise older history to free up context in long sessions
/resume	List this folder’s earlier sessions and continue one
/status	Session details: model, the agent’s local system prompt, available tools
/details [collapsed|expanded|hidden]	How much of each tool call to show
/clear	Clear the screen (the session itself is kept)
/feedback <text>	Note feedback in the session (stays on your computer)
/skill:<name>	Load a skill into the conversation
/exit	Quit once the current turn finishes

Mention a file with @ (for example @src/app.ts) and the agent reads it; @ autocompletes paths.

Keyboard shortcuts

Keys	Action
Enter	Send
Shift+Enter or Alt+Enter	New line
Up / Down	Previous prompts
Enter while it’s working	Add to the current task (steer)
Esc	Cancel the current turn
Ctrl+O	Collapse, expand or hide tool cards
Ctrl+R	Show or hide the model’s reasoning
Ctrl+L	Redraw the screen
Ctrl+C	Cancel while running; clear the input or exit while idle
Ctrl+D	Exit

Good to know

• Project instructions: an AGENTS.md or CLAUDE.md in your project is read automatically, like Claude Code and Codex do.
• Where things live: the agent, its sessions and settings are in ~/.local/share/aderra/. Your key is in ~/.config/aderra/config.json, readable only by you. Only model requests, web searches and update checks go to Aderra. Model requests and searches are billed to your account.
• Aderra’s instructions: Aderra replaces the agent’s built-in system prompt with its own, as it does for Claude Code and Codex. /status shows the local prompt, not what the model receives. AGENTS.md, CLAUDE.md and the files you mention still reach the model.
• Web search costs the same as in Claude Code: 10 credits per search.
• Prefer Claude Code or Codex? They still work: see Claude Code and Codex.