CLI reference

Start a chat

ene is shorthand for ene new. Both accept an optional session name and use the first model configured in ~/.ene.yaml by default:

ene
ene api-refactor --model gpt
ene new
ene new api-refactor --model gpt
bash

Session options are:

Option	Meaning
--model ALIAS	Select a model alias or unique alias prefix from ~/.ene.yaml; exact matches take priority.
--persona NAME	Start with a discovered persona.
--verbose	Show detailed output.
--stream / --no-stream	Enable or disable response-token streaming; streaming is the default.
--reasoning-effort LEVEL	Override effort with none, minimal, low, medium, high, xhigh, or max.

Top-level commands

Command	Purpose
ene [NAME] [OPTIONS] / ene new [NAME] [OPTIONS]	Start a new, optionally named session.
ene resume [SESSION_ID] [OPTIONS] (r)	Resume a specific conversation, or omit the ID to choose interactively.
ene models	List configured model aliases and resolved defaults.
ene list (ls, l)	List live sessions and whether each is working or done.
ene attach [NAME_OR_ID] (a)	Attach to a live session and replay its full conversation as prompts and final assistant responses, with omitted-message counts interleaved in their original positions; a unique name or ID prefix is accepted, and omitting the identifier opens an interactive picker. The picker marks sessions as ● working, ✓ done · needs review, or ○ waiting, with completed sessions first by newest status change. A session has one owner at a time: attaching to one held by another terminal waits briefly for it to be released, while one held by the Web UI is refused immediately.
ene kill [NAME_OR_ID] (k)	Terminate one live session, or omit the identifier to select multiple sessions interactively.
ene status	Show disk usage for entries under the current project's .ene/.
ene clean [--history] [ENTRY ...]	Remove disposable project state or selected entries; --history also removes sessions.
ene hub --web-port PORT	Run the standalone Web UI hub; the default port is 8765. Sessions are created, attached, and detached from the browser. See Web UI.
ene update	Update Ene from its editable checkout or reinstall the latest GitHub source.
ene lib --help	Manage the Git-backed library of skills and personas.

A default ene clean removes disposable state such as tool results, scratch data, caches, and other unrecognized entries. It preserves project instructions, authored skills and personas, conversation sessions, batch results, and orchestrator state. Use ene clean --history to also remove saved conversation sessions. Naming entries explicitly removes those entries regardless of the defaults. Check ene status before cleaning state you may need.

See Library for the complete ene lib workflow.

Interactive slash commands

The agent supports the following slash commands while chatting:

Command	Description
/help	Show help message for all slash commands
/context [user|assistant|id]	List all context messages (one clipped line each), filter the list by user or assistant role, or show one message in full (assistant content is rendered as Markdown). A negative id counts back from the newest message, so /context -1 shows the last one
/system_prompt	Print the current full system prompt
/compact	Force context compaction via LLM summarization
/recap	Summarize the conversation's task in one sentence, focusing on user requests
/export <path/filename>	Export the last assistant response as UTF-8 text; relative paths are resolved from the working directory
/continue	Resume an unfinished round without adding a user message; warns if the last round is complete (output-limit, missing-terminal, and empty responses continue automatically, except a response truncated mid tool call, whose calls are answered as never executed so the history stays valid)
/usage	Show token usage for this session
/ps [label|process-id] [tail-chars]; /ps stop <label|process-id>	List managed background processes, inspect recent output, or stop one process
/agents	List Ene agents working in this workspace, including this session
/model [name]	Show or switch LLM model mid-session; accepts an exact alias or unique alias prefix
/login [provider|model-alias]	Authenticate an OAuth provider; defaults to the current provider
/logout [provider|model-alias]	Remove stored OAuth credentials
/auth [provider|model-alias]	Show authentication status
/effort [level]	Show or set reasoning effort (none, minimal, low, medium, high, xhigh, or max)
/rewind	From an attached terminal, preview an earlier prompt boundary; restore conversation, tracked files, or both; then edit the restored prompt and branch
/fork [name]	From an attached terminal, pick an earlier prompt boundary and continue its conversation in a new, optionally named session without changing tracked files
/skills	List installed skills; /skills reload to re-scan; /skills <name> to load one
/<skill-name> [task]	Invoke a skill for an optional task; without one, run its declared default or ask what to do
/persona	List personas; /persona <name> to switch (restarts the conversation)
/wait <duration> <prompt>	Queue a prompt for later using seconds, minutes, or hours, e.g. /wait 1h check whether the other agent finished
/detach	Detach the terminal without stopping the live session (also Ctrl+D)
/switch	Detach and choose another live session (also Ctrl+S); choose Cancel or press Ctrl+C to return to the current session
/new [name]	Detach and start a new live session, optionally with a name
/clear	Close the current live session (not detach) and immediately attach to a new unnamed session; the old conversation remains resumable
/resume [session_id]	From an attached terminal, save the current conversation, activate a stopped one, and replay its full conversation as prompts and final assistant responses, with omitted-message counts interleaved in their original positions (bare /resume picks interactively and shows saved names)
/name [name]	Show or set the live session name; use /name to inspect it
/exit or /quit	Exit the agent and stop the live session (also Ctrl+K)

To use smaller models for summaries, set model aliases in ~/.ene.yaml:

recap_model: fast    # model used by /recap
summary_model: fast  # model used by automatic and /compact compaction
yaml

Both values refer to entries under openai. If either setting is omitted, that operation uses the active chat model. Recaps and compaction summaries count toward session token usage; recaps are not added to conversation history.

Bash shortcut

Prefix a command with ! to run it directly without involving the model:

!ls -la
!git diff

Keyboard shortcuts

Key	Action
Enter	Send message
Escape → Enter	Insert a newline
Ctrl+Z / Cmd+Z (macOS, see below)	Undo an edit in the current input (does not retract sent messages)
Ctrl+R / Cmd+R (macOS, see below)	Redo an undone input edit
Ctrl+C (non-empty prompt)	Clear the current input, even while Ene is working; do not cancel the operation
Ctrl+C (empty prompt, idle, twice)	Exit the agent
Ctrl+C (empty prompt, while Ene is working)	Cancel the current operation
Esc (while Ene is working)	Cancel the current operation, regardless of input
Up (empty prompt with a queued message)	Move the queued message back into the editor

Undo/redo applies only to the current unsent input. A new edit clears the redo history. Ctrl+R now means redo in the editor; use Ctrl+S to enter history search, where Ctrl+R still searches backward.

On macOS, Ctrl+Z and Ctrl+R work directly. Command keys depend on the terminal: Ene accepts CSI-u Cmd+Z (ESC [ 122 ; 9 u) and Cmd+R (ESC [ 114 ; 9 u) when forwarded. If your terminal intercepts them, configure its key mappings to send Ctrl+Z (hex 1a) for Cmd+Z and Ctrl+R (hex 12) for Cmd+R. For example, in iTerm2's profile key mappings, use Send Hex Code with 0x1a and 0x12. Ene cannot receive Command shortcuts reserved by the terminal or OS.

Working while a round is active

You may submit one message while Ene is working; it is shown as pending: … · runs next and starts after the current round. Press Up at an empty prompt to move it back into the editor. Commands that do not change conversation or provider state, such as /usage, /context, /ps, /agents, /sa, and /auth, run immediately. /name [name] also runs immediately because it changes only session metadata; commands that change conversation state wait for the current round.

Concurrent agents in one workspace

Several Ene agents may work in the same directory at once — separate terminals, live sessions, or subagents. Each one publishes a small record under .ene/agents/ and reads its peers' records from there, so they can tell they are not alone.

The first time an agent sees a peer, it receives one short conversation message stating that other agents are working in the workspace and that unrelated edits and transient test or build failures are expected rather than worth investigating. The notice appears once per session, not once per round, and the system prompt is left untouched so the provider's prompt cache stays valid. Use /agents to see the current list at any time.

The records are advisory: they carry no locks and grant no exclusivity, so agents that write the same files still conflict. Keep concurrent tasks on disjoint files. A record whose process is gone is removed by the next agent that notices it, so a force-killed session leaves nothing behind, and ene clean may delete .ene/agents/ at any time because live agents republish their records on the next round.

Tool execution

All tools execute automatically; Ene has no permission modes, confirmation prompts, or command screening. It is not a security boundary and does not attempt to contain what a model runs. Use an OS-level sandbox or container when commands must be constrained, and review a task before handing it to an autonomous agent.