Slash commands

In interactive mode, the following commands are available. When the line starts with /, a suggestion row appears below the input and narrows as you keep typing; press Tab to cycle through the completions. Commands that do not apply to the current session (for example /edit on a text provider) are not offered at all, and an unknown /word is sent as a normal message.

Command	Description
/file [path]	Attach a file (image, PDF, or text). With a path, attaches directly. With no path, opens a tabbed selector: "Attached" to remove attachments, "Add" to pick one from a directory browser.
/edit [prompt]	Edit a generated image by re-sending it as this turn's reference. With a prompt it takes the newest image (consecutive /edits iterate). With no prompt it opens a picker — preview on the left, every image this session generated on the right (newest first, labeled by its prompt, clickable path below) — and after you choose, type the prompt in the composer. Dedicated image providers (type: imagen / images) only.
/redo [prompt]	Re-send the last request: same reference images, same prompt unless you supply a new one. Bare /redo rolls the dice again (image models vary per call); /redo <reworded prompt> retries from the same canvas, so a rejected result never becomes the next input. Dedicated image providers only.
/new	Start a new session in the same run. The session you leave is saved and stays on disk — /session or iota resume <id> brings it back. The new one gets its own id, in the same place, and keeps what the chat runs under: the model and its settings (with where each came from, so one you set by hand stays yours), the system prompt, the tools and MCP servers, the directory, and whether the chat is being saved. It starts with an empty conversation, no attachments and the token counts at zero. The screen is not cleared: a dim New session <id> line marks where the new session begins (New session · not saved · /save keeps it for an unsaved chat). Background jobs keep running, and their results go to the new session. An unsaved chat a model has replied in asks Discard this unsaved chat and start a new one? first, and a session nothing has been said in yet is already new (Already a new session.). /clear is a hidden alias: it does the same and is not listed. A bot does not list it and refuses it: A bot's session never ends: to start over, run another agent or rename the bot.
/session	Tabbed selector over saved sessions: "Resume" to resume one, "Delete" to multi-select and delete others. A session open in another iota process is skipped by Delete, and a bot's session is neither offered nor deleted; inside a bot, /session is not a command.
/save [title]	Start persisting an ephemeral session (one started with --no-save or no_save: true): the whole backlog is written at once and auto-save continues from then on. An optional title is kept as-is; otherwise the model-generated one is used. Only offered while the session is ephemeral.
/model	Tabbed settings for the current session: "Model" picks the model (a combo box over the agent's choices — type to filter, or type a model name nothing lists and commit that), "Context" the context window, "Effort" the reasoning effort (default, low, medium, high, xhigh, max — passed to the provider verbatim, so a level the model doesn't support surfaces as an API error and you pick another), "Temperature" a slider (default omits the parameter), and a read-only "System" tab showing the system prompt exactly as sent — harness, your instructions and the overlay. Enter applies all tabs; only changed values are announced. Picking another model re-evaluates the five layered parameters against it — a window or effort the session inherited from the model it leaves is dropped, one you set here yourself is kept (see the five layered parameters). The rows are agents.<name>.choices: models: entries and inline provider:ids as they are, every provider:* listed at once — a source that cannot answer costs only its own rows, named in the panel's dim subtitle, and the input row stays open either way. A row reads provider:model (just the model when every choice is on one provider), with the models: entries that name it in brackets; typed text is read as -M reads it, so the name of a models: entry picks that entry, whether or not it is among the choices. The only annotation is (current); a row does not say what picking it does — the pick reports itself. A row on another provider moves the chat there: at once before the first reply; after a reply iota saves the chat, says how many replies carry payload the new provider will not replay, and asks Move this chat to <provider>:<model>? first — an unsaved chat is told to /save first. A bot refuses such a pick (see the choices). Image providers get their own tabs instead (see Image generation).
/compact [hint]	Summarize older history to free context; optional hint guides what to keep. While the summary is written, a Compacting context… spinner row — the one thinking uses — shows in the transcript, and the outcome replaces it; with nothing to summarize it says only Nothing to compact yet. Offered only while token accounting is live.
/export [file]	Export the conversation (saved sessions: the full on-disk log, so compaction never hides older rounds) to a single self-contained HTML file — the default — or Markdown with a .md/.markdown extension. With no argument, a selector picks the format and the filename is generated from the session title. Never overwrites an existing file.
/status	Show provider, model, context usage, last-turn token counts, the version, the mode (chat, agent or bot) and the session's title — an unsaved chat's too, (untitled) until it has one.
/jobs	Two tabs over the background jobs, refreshed every second: "Jobs" is a single-select list — id, elapsed, command; pick one for its full command, clock, pid, output file and the last lines of its output (Esc returns to the list) — and "Kill" a multi-select list that ends the jobs you tick (Space toggles, Enter kills; each one's notice lands as killed). Exists only while a job is running: a shell call that yielded after 20 s, or one started with background: true (see Background jobs).
/tools	One list of where the model's tools come from, refreshed while it is open: built in (the tools iota ships with) first, then each MCP server — its tool count and how it stands: connected, connecting…, down, needs login or failed, with the reason in a few words. Enter opens that source's page; Esc goes back to the list, on the row you left. A server's page shows its status, endpoint, transport, login, the names its tools go by (mcp__<server>__<tool>), its call timeout, the full reason it failed or went down (its stderr included), and its tools; Enter there reconnects the server in the background — the page shows it happening and how it went, and the chat gets one line, MCP <name> reconnected · N tools or why it did not. The built in page lists the tools and has nothing to reconnect. See Seeing what is connected.
/debug [on|off]	Request inspector. /debug on / /debug off toggle recording of API round trips (a debug marker appears in the status row while on); bare /debug opens the two-tab console — "Messages" (newest first, drill into a request/response pair) and the "Verbose" switch. Recording is off by default and MCP traffic is not recorded.
/skills [name [instructions]]	Bare /skills lists discovered agent skills — name, source (project/user), description, and any invalid skills that were skipped. /skills <name> runs one: its instructions (plus anything you add after the name) are sent as the message. Agent mode only; every discovered skill also shows up as a completion row.

Attached files are sent with your next message, then cleared automatically.

Keys

Key	What it does
Tab	Cycle through the slash-command completions in the suggestion row.
Esc	Cancel the innermost running scope — a streaming reply, or a running shell call. Background-job notices queued behind a turn you stop wait for your next message instead of starting turns of their own (see Rate limits and Esc).
Ctrl+C	Cancel the turn.
Enter	Send. You can keep typing while a reply streams; queued submits are sent in order. Ctrl+Enter sends too.
Ctrl+J	Insert a line break. Works in every terminal.
Alt+Enter, Shift+Enter	Insert a line break where the terminal reports the modifier (below); elsewhere they are a plain Enter and send.

A reply interrupted with Esc or Ctrl+C keeps its partial text in the history, marked interrupted.

Line breaks in the composer

Ctrl+J is the one that always works: it reaches iota as a different key from Enter in every terminal, with nothing to configure (unless a tmux plugin such as vim-tmux-navigator has bound C-j). The other two depend on what your terminal sends:

• Alt+Enter inserts a line break in Ghostty and tmux by default, and in Terminal.app with Use Option as Meta key on — without it, Option+Enter is a plain Enter and sends. Windows Terminal takes Alt+Enter for full screen by default.

• Shift+Enter inserts a line break where the terminal sends it as a key of its own; in most terminals it is a plain Enter and sends.

• Ghostty's default config sends Shift+Enter in a form iota cannot read, so it does nothing at all. Add this line to Ghostty's config file (ghostty +edit-config opens it) and reload the config or restart Ghostty:

  keybind = shift+enter=text:\n

  Shift+Enter then sends what Ctrl+J sends, and it works through tmux too.

• Ctrl+Enter stays a send. Most terminals send it as a plain Enter; in Ghostty's default config it, too, does nothing.

The composer grows with the rows a draft wraps to, and a queued message with line breaks shows each one as ⏎.

A reply that never comes

Esc and Ctrl+C always work, at any point of a request — press either to stop waiting. If you leave it, a provider that stops sending is not waited on forever: a streamed reply that sends no byte for 300 seconds fails with Response stalled and is not retried. What the turn finished before the stall is kept — its tool calls and their results, and the partial reply, marked as cut short — so sending your message again picks up from there; if it keeps happening, check the provider's url and the model. The 300 seconds is fixed, not a setting. A reply that fails after its text has reached the screen is not retried either: the failure is shown once, and what arrived is kept the same way, marked as possibly incomplete. A failure before any text arrives is still retried. While the request is out, the status row says Waiting for the model, and Waiting for the first token once the provider has answered; its clock counts the whole turn, from the moment you sent the message.

Rate limits and Esc

A finished background job normally wakes the model on its own. Two things stop that, so that iota does not keep starting turns nobody asked for:

• A rate limit or a usage cap. When a turn ends in Rate limited (429) — after at most three more tries — notices from jobs that finish afterwards wait instead of each starting a new attempt. Each leaves a dim line:

  ⏸ [background job b2 finished] — held until 17:42

  At that time iota tries once more with everything waiting (▶ Trying again with 2 held notice(s)): when the provider's retry-after says it may, otherwise a minute after the refusal, then less and less often, at most every 30 minutes. A try that is refused again leaves one dim line — ⏸ Still rate limited (429) — 2 notice(s) held; next try 18:12 — not another error block. Sending a message tries at once and takes the waiting notices with it; switching to another model or provider with /model tries at once there. Your own messages are never overtaken by these tries. The provider's own message, with its reset time if it gives one, is shown with the first refusal.

• Esc or Ctrl+C. Notices queued behind the turn you stopped, and any that arrive afterwards, wait the same way (⏸ … — held; it goes with your next message) and go with your next message. Nothing restarts until you speak.

A provider that asks to wait more than a minute — with a 429 or any other error — is not waited out in silence either: iota stops and holds until then (a day at the most, however long it asks), unless the provider explicitly insists on being retried. A message from another program is not held: it goes at once, as something you typed would.

Supported file types

Type	Extensions
Images	.jpg, .jpeg, .png, .gif, .webp
Documents	.pdf
Text	.txt, .md, .rs, .py, .js, .ts, .jsx, .tsx, .java, .c, .cpp, .h, .rb, .sh, .json, .yaml, .yml, .toml, .xml, .html, .css, .sql, .csv, .log, .ini, .cfg, .conf, and more