Commands & CLI

Slash commands

Run polyglav and type / to tab-complete commands. Use /help or /help <cmd> for details.

Command	Aliases	Description
/asks		List and answer parked asks from unattended runs (/asks list [all|pending|answered], /asks show <id>, /asks answer <id> <text>). Answering resumes the origin session. See config.md
/compact	/c	Summarize the conversation and trim the provider context
/config		Show, get, set, or unset config values (/config <key> <value>, /config unset <key>, /config --global <key> <value> for a global line). The listing appends each key's origin: (default), (global), or (local)
/connect		Connect a provider. /connect picks interactively from the known providers. /connect <name> presets a known provider's defaults and (re)enters its API key. /connect <url> detects a known host or creates a named custom provider (derived from the host, or /connect <url> <name> to name it). Tests the connection before saving, stores the API key + base URL in global providers.json, and writes provider/base_url into config - the model is picked separately with /model
/exit	/quit, /q	Save and exit
/focus		Show and switch the focused agent run. No args shows the current run, the run tree (each run with its id, session, and session id, plus a ↔ Switch to <role> marker), and the call log. Attach by <#run>, #<session_id>, session:<name>, root, or a session name, and navigate with parent, child [n], sibling [n], next, prev, or back. log [n] prints the focused run's buffered output (the last n lines, or all when omitted). Focus only navigates existing runs, it never creates a session. See architecture.md
/help	/h	Show available commands and tools (/help <cmd|tool> for detail)
/history		List the active session's turns as a numbered index: /history [n|all] [--thoughts [all]] [--run <target>]. Default last 10, --thoughts adds a dim thinking excerpt (all prints it in full), and --run reads another run's session without switching focus. See session.md
/jobs		Manage scheduled and durable jobs: list, status, show, add, add-supervisor, approve, reject, enable, disable, stop, remove, run. See jobs.md
/memorize		Summarize the active run into bounded memory: /memorize (active role), or /memorize <role\|team\|job> <name>. Writes .polyglav/memory/<scopes>/<name>.md. See config.md
/mode		Show or switch the agent mode (/mode plan = read-only, /mode build, or a custom mode)
/model		Show or switch the active model. /model <name> sets it on the current provider. /model <provider>/<model> switches provider and model together, approving the model on confirm
/models		List configured models, or probe a provider's available models. /models shows the approved-model history grouped by provider ((key) when the provider has a stored key). /models list [provider] probes a provider's advertised models live (default: current provider)
/plugins	/plugin	Manage plugins: list, enable, disable, install, update, uninstall
/print		Reprint a turn (or one part) in full: /print <n>[.<m>] [--full] [--run <target>]. Shows the turn metadata, prompt, thinking, each tool call with input/output, and the answer. Each part is capped by print_max_chars, and --full drops the cap. See session.md
/provider		Show or switch the active provider
/session		Show or switch the active session: /session new, /session load <name>, /session save. Saved-session catalog operations live under /sessions
/sessions		Manage the saved-session catalog: list, preview, delete, export <name> [out] (Markdown output, see session.md)
/skills		Manage skills: list, show <name>, new <name>, remove <name>. See skills.md
/teams		Manage teams: list (list <tag> filters), show <name>, new <name> [description], remove <name>, run <name> <task>. See teams.md
/thinking	/reasoning	Show or switch reasoning display and request (/thinking on streams reasoning dimmed under a blue - Thinking header, off shows only a spinner)
/tool		Run a tool directly (/tool <name> {"key": "value"})
/role		Show the active run's role and session. See roles.md
/roles		Manage roles: list (list <tag> filters), show <name>, new <name> [prompt], remove <name>. See roles.md
/unattended		Show or toggle unattended mode (/unattended on/off): no stdin at any depth, confirms auto-deny, human asks route to the lead or return without pausing. See config.md
/version	/v	Show the Polyglav version

/help renders commands with subcommands indented below and lists the allowed tools (policy- and mode-filtered, so plan mode hides write and exec tools) the same way under /tool. /tool with no arguments lists the same tools with their short descriptions.

Delegation is a normal tool: the lead agent proposes it, or you run it directly, and /tool delegate {"role": "researcher", "task": "..."} routes through the same tool policy. A configured role delegates without prompting (delegate defaults to allow. Set a role's delegate to ask to confirm), and a role outside the registry is denied. See roles.md and swarm.md.

CLI

usage: polyglav [-h] [--path PATH] [-v] {config,eval,export,fleet,jobs,mcp,models,plugins,run,serve} ...

Global:

Flag	Description
--path	Project path (default: current directory)
--version, -v	Print the installed version and exit

polyglav run

One-shot headless chat.

Flag	Description
--prompt, -p	Required. The prompt to send
--provider	Provider override (e.g. ollama, openai, groq)
--model	Model override (accepts a provider/model ref. Explicit --model auto-approves)
--approve-model	Approve the configured model ref without prompting
--base-url	Base URL override
--mode	Agent mode override (plan, build, or a custom mode)
--output	json (default) or text
--verbose	Print tool status and diagnostics to stderr
--session-id	Persistent session name (load or create)
--yes	Auto-approve tools that require confirmation
--no	Auto-deny tools that require confirmation (default)
--path	Project path

polyglav export

Export a saved session to Markdown (see session.md).

Flag	Default	Description
name	Required	Session name to export
--out	.polyglav/exports/<name>.md	Output file, - for stdout
--path		Project path

polyglav models

List configured models, or probe a provider's available models (mirrors /models).

polyglav models                          # approved-model history grouped by provider, `>` marks the active model, `(key)` a stored key
polyglav models list [provider]          # probe a provider's advertised models (default: current), exit 1 on failure

Flag	Default	Description
--path		Project path

polyglav eval

Tool-use evaluation harness: run task fixtures through the agent loop and report metrics. See eval.md.

polyglav eval --path <project> list              # list discovered fixtures
polyglav eval --path <project> run [--fixture <id>] [--provider P] [--model M]
polyglav eval --path <project> run --compare ollama,openai --output json

Flag	Default	Description
--fixture		Fixture id or substring (default: all)
--provider		Provider override
--model		Model override
--base-url		Base URL override
--compare		Comma-separated providers to compare
--output	table	table or json

polyglav config

Scoped, scriptable config management, with the same layers as /config (see config.md).

polyglav config get [key ...] [--show-origin]   # effective values, default all keys
polyglav config set <key> <value> [--global]    # JSON-parseable value, default local
polyglav config unset <key> [--global]          # drop a value from the selected scope
polyglav config reload                          # re-read the config files from disk

Flag	Default	Description
--global		Operate on ~/.config/polyglav/config.json
--local	(default)	Operate on the project .polyglav/config.json
--path		Project path
--show-origin		get only - append default/global/local

polyglav serve

HTTP JSON API server. See api.md. Also serves POST /mcp (MCP server) when the polyglav-core-mcp plugin is loaded.

Flag	Default	Description
--host	127.0.0.1	Bind address
--port	8787	Bind port
--path		Project path
--mode		Agent mode override (plan, build, or a custom mode)

polyglav mcp

Run polyglav as an MCP server over stdio (newline-delimited JSON-RPC). See mcp.md. Requires the polyglav-core-mcp plugin.

Flag	Default	Description
--path		Project path

polyglav jobs

Scheduled and durable jobs (cron / interval / one-shot), with retries, backoff, a human-in-the-loop approval gate, and recorded run history. See jobs.md.

Subcommand	Description
add	polyglav jobs add <name> --file jobs/<name>.md --cron "0 2 * * *" (or --interval N / --at ISO, with --prompt optional when --file given), plus --mode, --provider/--model, --role, --system-prompt, --tools-deny, --tool-permission, --retries, --backoff, --timeout, --require-approval, --approve-model, --approval auto
add-supervisor	polyglav jobs add-supervisor <name> [--interval N | --at ISO | --cron "expr"] [--task "text" | --file path] - scaffold an approved leader-typed supervisor job from the supervisor task template
approve	approve <name> - activate a job (or arm the next run for --require-approval jobs)
daemon	daemon [--tick 15] [--quiet] - scheduler loop, Ctrl-C to stop
edit	edit <name> - open the job's linked task file in $EDITOR (creates the template first)
enable / disable / stop	Toggle the enabled gate (stop = disable)
list	Table of jobs: schedule, status, next and last run
reject	reject <name> - send back to proposed and disable
remove	remove <name> - drop the definition (sessions are kept)
run	run <name> [--no-retry] [--verbose] - run now, apply retries, print the answer. --verbose streams the live turn. Exit 0 verified / 1 failed
show	polyglav jobs show <name> - definition plus full run history + last output
status	Runtime summary per job: fired count, last error, uptime, approval state

polyglav fleet

Supervise a fleet of scoped polyglav serve agents: ports, health checks, restart policy, config generation. See fleet.md.

Subcommand	Description
add	add <name> [--dir] [--port N] [--max-restarts N] - add an agent
config	config <name> --provider/--model/--role/--system-prompt/--mode/--tools-deny/--tool-permission - write only those keys into <dir>/.polyglav/config.json
down	Stop the supervised agents and the daemon if running
init	Scan immediate subdirectories holding .polyglav/config.json into the manifest
logs	logs <name> [n] [--follow] - tail an agent's .polyglav/logs/<name>.log
remove	remove <name> - remove an agent (stops it if running)
restart	restart [name|all] - stop, reset restart backoff, relaunch (default: all)
status	Live table: agent, enabled, port, pid, state, restarts, last error
up	up [--detach] - start agents. Ctrl-C = graceful down, --detach = background daemon

--path may be given before the subcommand (polyglav --path X fleet status) or after it (polyglav fleet --path X status).

polyglav plugins

Manage plugins headlessly. See plugins.md.

Subcommand	Description
disable	polyglav plugins disable <name> - remove a plugin from the plugins list (applies on next start)
enable	polyglav plugins enable <name> - add a plugin to the plugins list (applies on next start)
install	polyglav plugins install <git-url\|path> [--global] [--deps] - install a plugin
list	List installed plugins and their load status
uninstall	polyglav plugins uninstall <name> - remove a plugin
update	polyglav plugins update <name> - re-fetch from the recorded source

--path may be given before the subcommand (polyglav --path X plugins list) or after it (polyglav plugins --path X list).