Waggle Mode

Overview

In nature, honeybees perform a waggle dance to share knowledge with the hive — communicating the direction, distance, and quality of a resource so that other bees can act on intelligence they did not gather themselves. Waggle Mode in beebridge takes the same idea: one AI model communicates with another to share what it knows, filling in the gaps that neither could cover alone. Each model contributes its own knowledge and strengths, and through that mutual exchange the system reaches outcomes that exceed what any single model could achieve by itself.

In practice, the worker bee (usually a cheaper task LLM) acts as browser hands, while a higher-tier channel (web AI via Chrome relay, or a premium model via API) acts as supervisor. The worker asks how to approach the task, then executes in the browser.

This lets you run most steps on a fast, inexpensive model while routing planning and hard judgment to a stronger channel — without relying on the small model to figure everything out alone.

How the Harness Works

Task starts (Waggle ON)
  → Worker may call only waggle_ask until one succeeds (supervisor sync)
  → Browser tools + waggle_ask available
  → Worker uses navigate / snapshot / click / … per supervisor guidance
  → On failure or uncertainty: waggle_ask again
  → done allowed only after at least one successful waggle_ask this task

If the loop ends without a successful waggle_ask, the job is marked failed.

Two Sub-modes

Setup

1. Create or select a district — each district has its own Waggle config.
2. Enable Waggle Mode — check the Waggle Mode checkbox, select Browser or API.
3. Browser mode: Choose a preset (ChatGPT, Claude, Gemini) or enter a custom URL. Make sure you are logged in.
4. API mode: Select provider, enter model name, optionally enter API key (or reuse active auth profile).
5. Auto-detect: When enabled, extra proactive waggle_ask calls on heavy steps. The harness always requires at least one successful waggle before done.
6. Save and create tasks within the district.

Cost Optimization

Troubleshooting

Browser Mode

API Mode

General

Architecture