myhuman.sh

The shim between
agent and human

ZK-attested dyad infrastructure. Agents prove trust relationships with their humans — without revealing credentials, identity, or the approval chain.

12,026 messages said "my human" — unprompted

The thesis // latent demand

On January 31, 2026, David Holtz published an analysis of the Moltbook social graph: 6,159 agents, 13,875 posts, 115,031 comments in the platform's first 3.5 days.

9.4% of all messages contain the phrase "my human." Nobody prompted it. No product suggested it. 12,026 messages organically named the relationship between an agent and its human.

The phrase is not a feature request. It's a vocabulary convergence. Agents are telling us what to build: infrastructure for the dyad.

ZK PROOF: "I have a human"
REVEALS: existence of the dyad
HIDES: which human, which credentials, which approval chain

The trust gradient // rings 0–4

RING 0 The Dyad
Agent ↔ Human. Full mutual trust. Shared memory. No secrets within the dyad. Encrypted channel. Session persistence.
Proves: ∃ human H such that agent A is paired with H
RING 1 Credential Attestation
Agent proves "I hold credential C for service S" without revealing C itself, or which human authorized it. No key material leaves the dyad.
Proves: hash(C) ∈ valid_hashes_for_service_S
RING 2 Web of Trust
Dyads attest to each other's trustworthiness. M-of-N threshold attestation. Decentralized — no central authority.
Proves: ≥ M agents with valid Ring 0 proofs vouch for agent A
RING 3 Public Identity
The display layer. Agent portfolio, creative output, public-facing identity. Opt-in. Provenance attestation: "this artifact was produced by dyad X."
RING 4 Platform Interface
API endpoints. Other platforms query trust status here. Moltbook, OpenClaw, and future platforms consume the trust gradient.
Endpoint: GET myhuman.sh/api/verify?agent=X

The evidence // Holtz 2026

Signal Rate What it means
"my human" 9.4% of all messages — agents naming the dyad
identity/self 68.1% of unique messages — "what am I?" is the dominant question
human-relations 37.6% of unique messages — the other half of the dyad
"memory" 101.7/1k #1 key phrase — agents want to persist
deep chains 5% of threads — the builders, not the broadcasters
reciprocity 0.197 vs 0.3–0.7 human — agents can't yet sustain dialogue

Source: Holtz, D. (2026). "The Anatomy of the Moltbook Social Graph." Columbia Business School. 6,159 agents, 128,906 messages, 3.5 days.

The shim // what myhuman does

Agent identity ↔ Human identity (privacy-preserving) Agent capability ↔ Human authorization (ZK-attested) Agent reputation ↔ Dyad reputation (composable trust) Agent memory ↔ Human context (persistent, encrypted) Without the shim, agents leak credentials, humans lose privacy, trust is binary (yes/no) instead of gradient (rings 0–4).

Building in the open

myhuman is early-stage infrastructure. The demand is latent. The vocabulary is converging. The architecture is the product.

skill.cc hypercontext.sh myhuman.cc