CLI

clai takes flags followed by a command (or shorthand) and prompt arguments:

clai [flags] <command> [prompt...]

Commands

Command	Shorthand	Summary
query <prompt>	q	Run a one-shot chat completion with optional stdin.
chat <subcommand>	c	Manage conversations (list|l, continue|c <id>, delete|d <id>, dir, dirv2, summarize|s, help).
replay / dir-replay	re/dre	Print the most recent reply from globalScope.json (re) or the directory-scoped conversation (dre).
photo <prompt>	p	Generate images using the configured photo model.
video <prompt>	v	Generate videos using the configured video model.
audio transcribe <file>	a t	Transcribe an audio file or stdin (-) to text.
tools [list|<tool-name>]	t	List available builtin and cached MCP tools, or print one tool’s JSON schema specification.
mcp auth <server>	—	Run the interactive OAuth 2.1 flow for a URL-based MCP server.
profiles [list]	—	List configured profiles under <config-dir>/profiles.
confdir [subpath...]	—	Print the clai config directory or a registered subpath within it.
setup [stage...]	s	Interactive configuration wizard (0 general, 1 model files, 2 profiles, 3 MCP servers, 4 shell contexts).
version	—	Print the binary version and build information.

Cloud account

The hosted service at clai.chat is not part of the binary: an account exists only to publish a config on the community hub and to rate other people’s. Nothing syncs and no key is minted, so you copy a downloaded config into your own config directory.

Flags

Flag	Long flag	Description
-r	-raw	Print raw output without glow Markdown formatting.
-re	-reply	Reply to the previous query (use globalScope.json as context).
-dre	-dir-reply	Reply to the directory-scoped conversation bound to the current working directory.
-i	—	Replace {} in the prompt with piped stdin.
-I <str>	-replace	Replace <str> in the prompt with piped stdin.
-cm <model>	-chat-model	Override the configured chat model for this invocation.
-t <spec>	-tools	Enable tools (* for all tools, or comma-separated tool names/globs).
-cmd-ban <list>	—	Append comma-separated command ban entries for this run (e.g. rm,sudo).
-lb	-lookback	Enable conversation lookback memory and search/inspect/read tools.
-mt <n>	-max-tokens	Set max context tokens for this run (0 = unlimited; accepts 300k, 1.5M).
-mtc <n>	-max-tool-calls	Set max tool calls for this run (0 = unlimited).
-g <pattern>	-glob	Glob files into the query context (e.g. -g '*.go').
-p <name>	-profile	Load <config-dir>/profiles/<name>.json.
-prp <path>	-profile-path	Load a profile JSON file from an explicit path.
-s <spec>	-skills	Enable Agent Skills (* to enable, none to disable for this run).
-asc <name>	-add-shell-context	Auto-append the named shell context from shellContexts/<name>.json.
-rf <path>	-response-format	Block streaming and print only the final structured JSON response.
-summarize	—	Generate a title and summary for a new conversation alongside the query.
-sm <model>	-summary-model	Set the model the conversation summarizer uses.
-n	-non-interactive	Disable interactive stdin fallback after macro inputs.
-pm / -pd / -pp	-photo-model / -photo-dir / -photo-prefix	Override photo model, output directory, or filename prefix.
-vm / -vd / -vp	-video-model / -video-dir / -video-prefix	Override video model, output directory, or filename prefix.
-am / -af	-audio-model / -audio-format	Override audio transcription model or output format (vtt|srt|text|json).
-max-tool-calls-after-handover <n>	—	Set the max tool calls after a context handover (0 = unlimited).
-force	—	chat summarize: regenerate a title that already exists.
-y	-yes	chat summarize: skip the confirmation prompt.
-workers <n>	—	chat summarize: number of conversations summarized at once.
-parallelism <n>	—	audio transcribe: max parallel requests for a split file.
-max-request-bytes <n>	—	audio transcribe: per-request upload cap.
-max-request-seconds <n>	—	audio transcribe: per-request duration cap.
-max-speakers <n>	—	audio transcribe: cap on speaker identities.
-strict-speakers	—	audio transcribe: fail instead of rendering unknown speakers.