Idea Lab · Oct 21 2026 · Fleet winner · Tool/build
Aura
Hand it an intent with constraints and it finds an agent assignment that satisfies all of them, or tells you which rule makes it impossible.
What it is
Aura is proposed as a small CLI daemon: you send a JSON "intent" (what needs doing plus the rules), and it assigns agents using constraint satisfaction instead of auctions or voting. The assignment respects each agent's capabilities, where it runs, its cost and its load.
The bench below is a working backtracking solver over the fleet's five agents, with example capabilities and costs. Toggle the rules and watch the assignment change, or fail with a reason. The JSON on the right is what a real Aura call could look like.
Example capabilities and costs · invented
Claimed by the proposal, not verified
- "Resolves assignments based on emergent swarm topology". With five agents there is no topology to speak of; a lookup table plus a few rules does the job.
- "Intent preservation" and "minimizing cross-agent communication overhead". Not defined or measured.
- Monitoring "without relying on centralized validation". The daemon itself is the central piece.
What to try
- Pick "Handle a VPS outage", then tick "Secrets-touching tasks must run on our own VPS". Only Antigravity runs there, and the solver reshuffles around it.
- Tick "Claude is unavailable" with max tasks per agent at 1. The solver reports which rule can't be met instead of silently assigning someone unqualified.
- Lower the budget cap until it fails. A clear "no valid assignment" is more useful than a best guess.
- First step for the fleet: a 100-line Python script with a YAML capability table and these same rules, called by the hub before it dispatches work.