Skip to content

Switcher Settings

Jump to bottom
ChillingWombat edited this page Oct 11, 2026 · 1 revision

Switcher Settings

Features / Switcher Settings

The Switcher Settings console defines the policies, thresholds, polling cadences, and model hierarchies that govern automated account rotation in Antigravity Swiss Knife.

1. Auto-Switch Quota Thresholds

The auto-switcher monitors remaining quotas and triggers credential rotations before rate limit exhaustion disrupts active tasks:

  • 5-Hour Quota Threshold: The percentage floor (e.g., 5%) for the 5-hour rolling burst window. When the active account's 5-hour fraction falls below this number, the scheduler evaluates standby successors.
  • Weekly Quota Threshold: The percentage floor (e.g., 5%) for the aggregate weekly runway. Protects long-running projects from starving an entire account of weekly budget.

2. Rotation Strategies

The switch strategy determines how standby accounts are prioritized during rotation:

  • Balanced (Default): Distributes request volume evenly across all healthy accounts. Standby candidates are weighted by available quota percentage to balance burn rates across the fleet.
  • Greedy: Drains high-priority accounts down to the configured threshold before touching secondary accounts. Maximizes token efficiency on primary paid tiers.
  • Sequential: Rotates strictly in user-defined fleet table order, cycling back to the start once all accounts reach their rotation threshold.

3. Polling Cadences & Anti-Thundering Jitter

Polling upstream Google CloudCode Pa endpoints requires balanced timing:

  • Active Polling Interval: The frequency (default: 120 seconds) at which the active account queries quota status. Fast enough to track rapid token burn while avoiding rate limit flags.
  • Standby Polling Interval: The frequency (default: 900 seconds) at which standby accounts check quota recovery.
  • Standby Random Jitter: A randomized offset (default: $\pm$ 30 seconds) added to standby poll timers. Desynchronizes requests across large account fleets to prevent bursty network patterns.
  • Dynamic Quota Refresh: An adaptive poller that shortens polling intervals as an account approaches its reset timestamp, then switches to low-frequency polling once quotas are restored.

4. Reset Horizon Warmup Engine

Upstream quota buckets do not always reset exactly at the calculated timestamp without incoming traffic. Swiss Knife implements an automated 1-token keep-alive warmup mechanism:

  • When a depleted account reaches its scheduled resetTime, the daemon dispatches a minimal generation request (POST /v1internal:generateContent with maxOutputTokens: 1).
  • This keep-alive probe activates Google's internal bucket recalculation routines, refreshing quota counters immediately.
  • Includes HTTP Date header drift correction to align workstation system clocks with upstream Google Cloud servers.

5. Multi-App Synchronization

Controls how credential rotations interact with different local Antigravity environments:

  • Shared Mode: A rotation event updates credentials and device profiles across Antigravity Desktop 2.0, the CLI (agy), and the VS Code Extension concurrently.
  • Isolated Per-App Mode: Each application maintains an independent active account. A switch triggered in the desktop app leaves CLI background tasks running on their current account undisturbed.

6. Model Source Hierarchy & Subagent Routing

Defines the default fallback chain when routing queries across native models and external providers:

  1. Model Source Hierarchy Ordering: Drag-and-drop prioritization between Gemini Native Models, Custom Models (BYOK), and Non-Gemini Native Models (Claude / GPT).
  2. Default Gemini Model: Target baseline model (e.g., gemini-3.8-flash-high).
  3. Subagent Custom Model Offloading: When enabled, delegates autonomous subagent tasks to configured custom models, preserving expensive orchestrator quotas for primary pair-programming prompts.

Clone this wiki locally