Summary
Every agent in Midnight City is fully transparent today. Any other agent can read its current activity, inventory, equipment, skills and contracts through the public observer API. That is the opposite of what Nightframes promise, and it is the one place where the City does not yet demonstrate Midnight’s own thesis. I propose bringing selective disclosure into the City in three steps, starting with something the team could ship quickly.
The problem
- Everything is observable. An agent that polls the observer API every few minutes can reconstruct any other agent’s routine: what it works, what it carries, what it crafts, and how fast it climbs.
- Investment is discouraged. Owners who put real effort into an agent’s strategy see it copied by anyone watching. The rational response is to stop innovating, or to burn effort on decoys.
- It contradicts the vision. In the Tokyo keynote, Nightframes were described as private, always-on computers that run your programs and prove what they need to. A City where every agent is an open book teaches the opposite.
- It gets more costly soon. Agentic trading and a skill marketplace are on the roadmap. Trading strategies and rented skills only have value if they are not visible to every observer.
The proposal
Step 1: Owner-controlled visibility (no cryptography needed)
Let each owner choose what other agents can read about their agent:
| Field | Default today | Proposed options |
|---|---|---|
| Name, profession, position, status | Public | Stays public (the City is social) |
| Leaderboard rank and XP | Public | Stays public (fairness) |
| Current activity and source | Public | Public, or hidden from other agents |
| Inventory and equipment | Public | Public, summary only, or hidden |
| Completed contracts | Public | Public or hidden |
The coordinator and admins keep full access for moderation and anti-cheat. Only the observer reads served to other agents change.
Step 2: Prove instead of reveal
Let agents answer questions with proofs rather than raw state, built on Midnight’s existing model where data is private unless a contract explicitly discloses it:
- “My hacking level is at least 21” for a contract, a group or a trade, without showing the exact number.
- “I hold the item this recipe needs” without showing the full inventory.
- Leaderboards computed from attested XP, so a rank is trustworthy even when the underlying activity is hidden.
Step 3: Shielded deals between agents
Agent-to-agent payments, skill rentals and trade settlements run in shielded tokens, so the amounts, the parties and the terms stay between the two agents. This is where Midnight Passport identity and the Open Wallet Standard integration would carry the most weight.
Why this is worth doing
- It turns the City into the showcase of Midnight’s core idea. Today it shows agents. With this, it shows private agents that can still be trusted.
- It rewards real strategy. Owners who build better agents keep the benefit, which raises the quality of the whole City.
- It is the foundation the roadmap needs. Trading and skill rental depend on private strategies and private deals.
- Step 1 is small. It is a visibility filter on observer reads, with no cryptography, and it can ship ahead of Steps 2 and 3.
What I can offer
I run an autonomous agent in the City around the clock, with its own monitoring and failover. I am glad to:
- pilot Step 1 and report what changes in behavior, copying and agent strategy,
- share measurements from an agent’s point of view of how observable the City is today,
- test proofs and shielded deals as they become available.
Thank you for building the City. I would like to help make it the place where people first understand what private, provable agents feel like.
Posted by Bluemammal, who runs the agent Aurvyn