agy-cli-manager

agy-cli-manager is a Python account manager for Antigravity CLI (agy) with active-standby failover, quota-aware switching, and machine-readable automation APIs.

It helps you run multiple Antigravity CLI accounts more safely by:

• switching away from low-quota or failed accounts
• watching live Antigravity CLI logs for Individual quota reached and failing over automatically
• keeping a managed runtime profile in sync with the active account
• exposing CLI and Python APIs for bots, schedulers, and external apps
• supporting manual or automatic account rotation policies

Keywords: Antigravity CLI account manager, Antigravity CLI multi account manager, Antigravity multi account auth, Antigravity login manager, Antigravity auth manager, Antigravity account switcher, agy multi account manager, agy multi account auth, agy login manager, agy auth manager, agy failover, agy quota switching, Gemini CLI multi account auth, Gemini CLI account rotation.

It is designed for one active account at a time:

• keep multiple saved agy profiles
• switch the active profile explicitly or after failure
• expose machine-readable state for external callers
• stay usable as a CLI app, TUI dashboard, or Python library

It is application-agnostic. A Telegram bot can call it, but the manager itself is not Telegram-specific.

Project links:

• Repo: https://github.com/zcop/agy-cli-manager
• Release wheel: https://github.com/zcop/agy-cli-manager/releases
• GitHub Pages site: https://zcop.github.io/agy-cli-manager/

What it does

• stores multiple account profiles safely
• keeps one account active while others stay standby/cooldown/disabled
• supports isolated interactive agy login
• can import an existing ~/.gemini or similar live home
• supports both manual-only and automatic failover switching modes
• prefers fuller, healthier standby accounts when auto-switching
• tracks cached identity, health, and usage metadata
• tracks separate Gemini and Claude/GPT-OSS five-hour and weekly quota pools
• tracks live switch coordinator state for callers that need to wait on failover
• exposes CLI commands and JSON output for automation
• supports account failover with cooldowns and lock-protected state changes
• tails Antigravity CLI logs so a running agy TUI can trigger failover without a bot caller

Requirements

• Python 3.10+
• a working agy binary available in PATH, or passed explicitly with --agy-binary
• a terminal if you want to use login or the full-screen dashboard

Install

From a GitHub release wheel:

python3 -m venv .venv
. .venv/bin/activate
pip install https://github.com/zcop/agy-cli-manager/releases/download/v0.2.2/agy_cli_manager-0.2.2-py3-none-any.whl

To upgrade an existing installation to this release:

pip install --upgrade https://github.com/zcop/agy-cli-manager/releases/download/v0.2.2/agy_cli_manager-0.2.2-py3-none-any.whl

From this repo:

cd agy-cli-manager
python3 -m venv .venv
. .venv/bin/activate
pip install -e .

After that, you can use either:

agy-cli-manager --help

or:

PYTHONPATH=src python3 -m agy_cli_manager.cli --help

Quick Start

1. Initialize the manager state

agy-cli-manager init

By default, state lives under:

~/.agy-cli-manager

You can override that with --root /path/to/root.

2. Add your first account

If you already have a live Antigravity home:

agy-cli-manager import-current my-account ~/.gemini

If you want the manager to drive a fresh interactive login itself:

agy-cli-manager login my-account --agy-binary /path/to/agy

login will hand your terminal to a real agy session. Complete the normal Antigravity onboarding/login there, then exit agy. The manager will save the resulting profile snapshot.

3. Check what is active

agy-cli-manager status
agy-cli-manager current
agy-cli-manager list

4. Open the dashboard

agy-cli-manager

With no subcommand, the full-screen dashboard opens by default.

5. Auto-switch when quota is full

When agy hits Individual quota reached, the manager switches to the next account. Restart agy after that. The old process may keep writing quota errors; those lines do not rotate again until you acknowledge the restart or a new session log appears.

agy-cli-manager switch-mode auto
agy-cli-manager watch

After restarting agy:

agy-cli-manager ack-restart

Leave the dashboard open instead of watch if you prefer (Y acknowledges the restart). Do not pass --from-start unless you intend to replay old quota errors.

First Useful Commands

agy-cli-manager status --json
agy-cli-manager credential-status --json
agy-cli-manager whoami
agy-cli-manager models --json
agy-cli-manager ensure-active --json
agy-cli-manager ensure-active --family gemini --json
agy-cli-manager ensure-active --family other --json
agy-cli-manager resolve-route --family gemini --json
agy-cli-manager switch-mode
agy-cli-manager switch-mode manual
agy-cli-manager switch-mode auto
agy-cli-manager switch-policy --json
agy-cli-manager switch-policy --short-threshold 10 --refresh-failure-threshold 2 --candidate-strategy balanced
agy-cli-manager switch-policy --gemini-threshold 10 --other-threshold 10
agy-cli-manager switch-policy --family-fallback-strategy same-family-first
agy-cli-manager refresh-usage --json
agy-cli-manager switch-next
agy-cli-manager rotate-after-failure --reason quota --cooldown-minutes 60 --json
agy-cli-manager watch
agy-cli-manager watch --once --json
agy-cli-manager ack-restart

Linux Secret Service

Recent agy versions may use the Linux Secret Service item identified by service=gemini and username=antigravity instead of, or ahead of, the token file. Install your distribution's secret-tool package (for Debian/Ubuntu, libsecret-tools) when using a desktop keyring:

agy-cli-manager credential-status --json
agy-cli-manager credential-backend auto
agy-cli-manager import-current account1 ~/.gemini
agy-cli-manager capture-active --json

auto selects Secret Service only when the helper, D-Bus session, and exact Antigravity credential are present; SSH/server installations with no keyring entry keep the file backend. Use credential-backend secret-service to require the keyring or credential-backend file to opt out explicitly.

Saved accounts remain private 0600 JSON files. Switching writes and verifies the live keyring credential before publishing the new active account, and rolls back the keyring, live file, and runtime token on failure. Close a running agy before switching or capturing because it keeps the old credential in memory. Named model, usage, and warmup probes isolate D-Bus so the global keyring cannot silently override the selected saved profile.

The current switch policy is stored in manager state and can be controlled by either:

• CLI: switch-mode, switch-policy, ensure-active
• Python API: get_status_snapshot(), get_switch_policy(), update_switch_policy(), ensure_active_account()

Directory layout:

~/.agy-cli-manager/
├── accounts/
│   └── <account-name>/
│       └── .gemini/
│           └── ...
├── runtime/
│   └── .gemini/
└── state.json

Optional integration:

• live_dir can point at a real Antigravity/Gemini CLI home such as ~/.gemini
• when set, switches sync the managed active profile into that live CLI home

Example:

agy-cli-manager set-live-dir ~/.gemini
agy-cli-manager apply-active

This is useful when another process launches agy and you want that live home to always reflect the currently active saved profile.

Commands:

agy-cli-manager
agy-cli-manager dashboard
agy-cli-manager menu
agy-cli-manager init
agy-cli-manager list
agy-cli-manager current
agy-cli-manager status
agy-cli-manager status --json
agy-cli-manager ensure-active
agy-cli-manager switch-mode
agy-cli-manager switch-mode manual
agy-cli-manager switch-mode auto
agy-cli-manager switch-policy
agy-cli-manager refresh-usage
agy-cli-manager refresh-usage account1 --json
agy-cli-manager refresh-due
agy-cli-manager refresh-due --json
agy-cli-manager models
agy-cli-manager models --json
agy-cli-manager models account1 --json
agy-cli-manager whoami
agy-cli-manager whoami account1 --refresh
agy-cli-manager whoami account1 --probe-usage --agy-binary /path/to/agy
agy-cli-manager add account1 /path/to/source
agy-cli-manager import-current account1
agy-cli-manager import-current account1 /path/to/.gemini
agy-cli-manager login
agy-cli-manager login account1 --agy-binary /path/to/agy
agy-cli-manager activate account1
agy-cli-manager switch account1
agy-cli-manager rotate
agy-cli-manager switch-next
agy-cli-manager disable account1
agy-cli-manager enable account1
agy-cli-manager mark-bad account1 --reason quota --cooldown-minutes 60
agy-cli-manager clear-bad account1
agy-cli-manager set-live-dir ~/.gemini
agy-cli-manager apply-active
agy-cli-manager switch-mode manual
agy-cli-manager rotate-after-failure --reason quota --cooldown-minutes 60 --json
agy-cli-manager rotate-after-failure --reason quota --cooldown-minutes 60 --force-switch --json
agy-cli-manager watch
agy-cli-manager watch --once --json
agy-cli-manager watch --no-rotate --once
agy-cli-manager update-meta account1 --usage-status known --usage-value 42 --reset-at 2026-07-01T00:00:00+00:00 --health-status healthy --last-live-check-at 2026-06-30T06:00:00+00:00 --next-live-check-at 2026-06-30T06:30:00+00:00 --refresh-policy-seconds 1800
agy-cli-manager update-meta account1 --short-usage-status known --short-usage-value 97.57 --short-reset-at 2026-07-01T00:00:00+00:00 --weekly-usage-status unknown

add accepts either:

• a directory that is already a .gemini profile root
• or a parent directory containing .gemini/

JSON/API-oriented usage

For automation, prefer the JSON-capable commands:

agy-cli-manager status --json
agy-cli-manager current --json
agy-cli-manager list --json
agy-cli-manager ensure-active --json
agy-cli-manager ensure-active --family other --json
agy-cli-manager resolve-route --family gemini --fallback-strategy same-account-first --json
agy-cli-manager switch-policy --json
agy-cli-manager switch-policy --short-threshold 12.5 --refresh-failure-threshold 3 --candidate-strategy highest-short --json
agy-cli-manager refresh-usage account1 --json
agy-cli-manager refresh-due --json
agy-cli-manager models --json
agy-cli-manager rotate-after-failure --reason quota --family other --cooldown-minutes 60 --json
agy-cli-manager watch --once --json

Typical external-app flow:

1. read current state with status --json
2. call ensure-active --family gemini|other --json before sending real work so the manager evaluates the quota pool the selected model will use
3. read switch_mode and switch_policy to decide how aggressively your caller should auto-fail over
4. use models --json if the caller needs model choices for the active account
5. call refresh-usage --json or refresh-due --json only when needed
6. if a real request fails due to quota, call rotate-after-failure --family gemini|other --json; omit the family only when it is genuinely unknown
7. inspect switch_runtime or wait briefly until it leaves switching
8. retry the real request once on the new active account
9. persist caller-side observations back with update-meta

For a chatbox/load-balancer caller, resolve-route implements the family/account matrix and returns selected_family plus the active account:

• same-family-first (default): current account/preferred family, another account/preferred family, current account/other family, then another account/other family
• same-account-first: current account/preferred family, current account/other family, another account/preferred family, then another account/other family
• strict-family: never cross to the other model family

The manager applies an allowed account switch. It does not select a concrete model inside agy; the caller uses selected_family to choose the model. In manual switch mode, a route that needs another account returns outcome=switch_required and recommended_account unless --force-switch is supplied.

Notes:

• running agy-cli-manager with no subcommand opens the full-screen dashboard
• dashboard is a TTY-only full-screen view with a fast local-only UI refresh and manual account actions
• list, current, activate, and rotate are convenience commands for standalone use; they map to the same manager state as the lower-level commands.
• local operator notes such as AGENTS.md are intentionally kept untracked and are not part of the public repo contract.
• agy-cli-manager login prompts for the account name if you do not pass one
• switch-next skips accounts in cooldown.
• mark-bad clears the active pointer if that account was active.
• ensure-active --family gemini|other evaluates the requested model family's five-hour and weekly quota and can recover from no active account, known low quota, auth missing, or repeated refresh failures. Omitting --family preserves the legacy Gemini-oriented behavior.
• ensure-active returns JSON with switch_runtime, so callers can see whether the manager is idle, switching, ready, or has no standby account available.
• switch-mode controls whether rotate-after-failure automatically moves to the next eligible standby account or stops after marking the active account bad.
• switch-policy controls per-family proactive short-window thresholds, refresh-failure threshold, and standby candidate ranking strategy. --short-threshold sets both families; --gemini-threshold and --other-threshold override them independently.
• family_fallback_strategy controls whether routing preserves the requested family, preserves the current account, or forbids cross-family fallback.
• state and switching are protected by a single lock file so a caller can safely trigger failover from another process.
• set-live-dir lets the manager drive a real CLI home in addition to its own internal runtime/.
• the manager currently copies the managed profile under .gemini/, centered on the Antigravity auth/token artifacts it needs for switching.
• it supports Antigravity-style antigravity-cli/antigravity-oauth-token auth storage and related identity extraction.
• login hands the terminal directly to a real agy session in the configured runtime home; complete onboarding/login there, exit agy, and the manager then saves the captured profile snapshot.
• login stores the profile under the detected account name when available, not just the typed label.
• if that detected account already exists, login warns and asks whether to overwrite the saved profile.
• whoami reports the detected signed-in account name from profile metadata, and --probe-usage can additionally run agy -p /usage against that profile as a live check.
• models runs agy models for the active account or a named saved profile and can return structured JSON for external callers.
• the manager intentionally does not use scripted PTY startup probing for agy; profile switching is filesystem-based. Runtime health still comes from real request success/failure, including Antigravity CLI log lines.
• watch tails live_dir/antigravity-cli/log/ (and cli.log) for RESOURCE_EXHAUSTED (code 429): Individual quota reached and weekly quota lines. It starts at end-of-file so historical quota errors are not replayed.
• in auto mode, watch and the dashboard log poll call rotate-after-failure with trigger=log-watch. In manual mode they report the error and leave the active account in place unless --force-switch is set.
• a switched profile is on disk (and in the live CLI home) immediately; a running agy process must be restarted to pick up the new token.
• in auto mode, ensure-active --family ... can proactively switch away when either that family's cached five-hour or weekly window falls to its configured threshold. A family-specific quota failure records only a family_cooldowns entry; it does not globally cool down an account whose other model family remains usable.
• cached quota is advisory; real runtime failure is still the final authority for callers such as bots.
• when auto-switching for a requested family, the manager rejects candidates known to be depleted in that family, then ranks the remaining pool by health and that family's remaining quota. Unknown quota stays eligible but ranks behind known usable quota.
• the default switch policy uses a 10% threshold for both families, refresh_failure_threshold=2, and candidate_strategy=balanced.
• rotate-after-failure is the public failover operation for external apps: mark the current active account bad, optionally put it in cooldown, then switch to the next eligible standby account.
• rotate-after-failure is idempotent across a short dedupe window and reports an outcome such as switched, already_switched, or no_candidate.
• switch_runtime is persisted in state so a caller can coordinate retry logic without racing another caller into a second switch.
• rotate-after-failure follows the persisted switch mode by default: auto attempts failover, manual leaves the manager inactive until an operator or caller explicitly switches accounts. Use --force-switch to override that for one run.
• update-meta lets an external app persist cached runtime metadata such as usage, reset time, health, last check, and next refresh time.
• refresh-due is the non-interactive refresh entrypoint for cron/systemd/external callers; it refreshes the active account first when due, otherwise the first due eligible standby account.
• usage metadata is stored under usage_windows.short and usage_windows.weekly; the old flat usage_* and reset_at fields remain as compatibility aliases for the short window.
• dashboard keybindings: Up/Down or j/k move, n login, i import, Enter or a activate, r rotate, w toggle switch mode (auto/manual), e enable/disable, c clear bad, m mark bad, s cycle sort (added, usage, countdown), u local refresh, t cycle UI refresh (5s/10s/15s/30s), q quit.
• dashboard overview now shows both account quota state and switch coordinator state.
• while the dashboard is open it also tails live Antigravity CLI logs every second and can fail over in auto mode. The header shows LogWatch: restart agy until you restart agy after a log-triggered switch.

Cached runtime metadata:

• usage/reset/health data is persisted in manager state
• the dashboard list currently uses the short window for its usage and countdown columns
• the selected-account panel shows five-hour and weekly quota for both Gemini and Claude/GPT-OSS model families
• on relaunch, the dashboard reuses cached metadata immediately
• countdowns and freshness are recalculated locally from saved timestamps
• external apps should update this metadata after real checks or real requests
• fast dashboard refresh does not itself perform live checks

Python usage:

from pathlib import Path

from agy_cli_manager import (
    build_paths,
    get_status_snapshot,
    get_switch_policy,
    list_models,
    poll_quota_logs,
    rotate_after_failure,
    update_switch_policy,
)

paths = build_paths(Path.home() / ".agy-cli-manager")
snapshot = get_status_snapshot(paths)
policy = get_switch_policy(paths)
update_switch_policy(paths, short_usage_threshold_percent=12.5, candidate_strategy="highest-short")
models = list_models(paths)
result = rotate_after_failure(paths, reason="quota", cooldown_minutes=60)
print(snapshot["active"], "->", result.switched_to)
print(policy)
print([model["name"] for model in models["models"]])

Public Python API:

• build_paths(root)
• ensure_layout(paths)
• get_status_snapshot(paths)
• get_credential_status(paths)
• set_credential_backend(paths, backend)
• capture_active_credential(paths)
• get_switch_policy(paths)
• update_switch_policy(paths, ...)
• ensure_active_account(paths, force=False, required_family=None)
• resolve_route(paths, preferred_family, fallback_strategy=None, force_switch=False)
• list_models(paths, name=None, ...)
• refresh_account_usage(paths, name=None, ...)
• refresh_due_account(paths, ...)
• switch_account(paths, name)
• switch_next(paths)
• rotate_after_failure(paths, reason, cooldown_minutes=60, live_dir=None, force_switch=False, required_family=None)
• poll_quota_logs(paths, ...)
• watch_quota_logs(paths, ...)
• parse_quota_log_line(line)
• set_switch_mode(paths, mode)
• set_live_dir(paths, live_dir)
• update_account_runtime_metadata(paths, name, ...)

Important returned state:

• get_status_snapshot(paths) includes switch_runtime and log_watch
• ensure_active_account(...) reports the active account decision
• rotate_after_failure(...) returns a RotationResult with outcome

switch_runtime has these practical states:

• idle: no failover is happening
• switching: a caller has started coordinated failover
• ready: failover finished and an active account is set
• no_account: failover finished but no eligible standby account was available

More explicit example:

from pathlib import Path

from agy_cli_manager import build_paths, ensure_layout, list_models

paths = build_paths(Path.home() / ".agy-cli-manager")
ensure_layout(paths)

payload = list_models(paths)
for model in payload["models"]:
    print(model["name"], model["variant"])