One brief goes in. An AI planner breaks it into however many tasks it actually needs, spreads them across whichever model providers you've connected, and assembles what comes back into one result.
Not a chatbot with extra steps. Four real stages, each one an independent API call you can inspect after it finishes.
One sentence or ten. No form to fill in about "roles" or "agents" — just the outcome you're after, in your own words.
One model reads the brief and returns a flat list of independent subtasks — as few as three, as many as twenty-four, whatever the brief actually calls for. Nothing is hardcoded to a fixed number of "agents."
Each subtask goes to whichever provider you've connected — Groq, OpenRouter, Gemini — round-robined so no single backend carries the whole load. Workers run in parallel, batched into clusters so results stay coherent instead of drowning one synthesis call in noise.
A final pass reads every worker's output, resolves overlaps, and returns a single coherent piece — not a list of disconnected fragments you have to stitch together yourself.
HIVE exists for briefs bigger than a single prompt can hold — where the work benefits from being split, run in parallel, and reassembled.
A three-item brief gets three workers. A sprawling one gets twenty. The planner decides the shape — you never pre-configure how many "agents" a task deserves.
Work is round-robined across every backend you've connected. Skip a provider you don't have a key for and the hive simply routes around it.
Workers are grouped into batches before a final pass combines them — so twenty-four outputs don't get flattened into mush by one overloaded synthesis call.
Tap any worker, any cluster, or the core result mid-run to read exactly what it produced. Nothing happens in a black box.
There's no backend here to trust. Every request goes straight from your device to the provider you typed the key in for.