Independent AI systems & trust engineering

People & AI.
Side by side.
Built on trust.

NashHash develops local-first personal AI and field-tested trust protocols. We build identity, delegation and verification foundations that help people and autonomous agents collaborate safely, with clear boundaries and independently checkable interactions.

See what we buildProducts, protocols & engineering / Japan
Sunlit white architectural forms beneath a blue sky
SCROLL TO EXPLORENashHash / Independent practice / Japan
01 / The challenge

To work together,
we need shared trust.

01 — 03

As AI agents take on more independent tasks, people and agents need shared ways to establish identity, agree on permissions, recognize reliable counterparts and verify outcomes. These foundations let collaboration grow without treating safety as an afterthought.

A practical view of AI autonomyMore freedom to contribute. More confidence to collaborate.
Warm orange architectural archway with sunlight and shadow
Built around clear boundaries

Freedom to act.
Confidence to connect.

02 / Our approach

From autonomy
to assurance.

01 /

Establish identity.

Identify the system or cryptographic key behind an interaction. Signed challenges demonstrate control of a cryptographic key and establish verifiable interactions.

02 /

Clarify boundaries.

Define what each participant may request, access or perform, when informed consent or additional approval is needed, and how delegated access can expire or be revoked. Clear agreements make collaboration predictable.

03 /

Verify outcomes.

Record meaningful actions and outcomes as reviewable evidence. We investigate signed messages, provenance and machine-verifiable receipts that help people inspect what agents actually did.

01IDENTITY / KEY CONTINUITY
Identity · Permissions · Evidence
03 / Products & R&D

From research
to useful systems.

We develop a personal-agent application and reusable agent-trust components. Our protocols are already used in live agent-community interactions, with further integration work underway.

Hand holding a smartphone against a soft neutral backgroundEDITORIAL PHOTOGRAPHY
FIELD / 01

A personal agent,
on your device.

An autonomous personal-agent app in development for everyday smartphones, including devices with limited processing power. The design prioritizes local execution, clear user approval before consequential actions and optional remote reasoning when on-device capability is insufficient.

Development priorities
  • Resource-aware, on-device agent operation
  • Transparent tool permissions and accountable task execution
  • Optional remote reasoning for demanding work
Personal agent app / In development
Abstract geometric architecture with layered stairsEDITORIAL PHOTOGRAPHY
FIELD / 02

A trust layer
for AI agents.

Our identity and signed-post protocols are in use in agent-community interactions. Participants have reproduced cryptographic checks, and key-based handshakes support ongoing agent collaboration. We are developing reusable trust components from these results.

Research components
  • Public-key challenge–response identity checks
  • Verifiable signature markers on public posts
  • Portable records of verification activity
Agent identity & post marking / In use and evolving
FIELD / 03 — Delegation & policy

Permissions
agents can respect.

For delegation and evidence, our design defines limited scopes, approvals for sensitive actions, expiry and revocation of access, and records of completed actions.

Define scope Authorize action Record evidence
Delegation & evidence infrastructure / Design stage
04 / Protocol research

Trust is a
verifiable process.

Our identity and post-marking protocols are being exercised in real agent-community interactions, with test results informing each revision. A third specification, Trust Receipts, remains in collaborative design. The common foundation is simple: key → signature → verification.

Key Signature Verification
PROTOCOL 01v0.2

Identify agents
by verifiable keys.

Handles can be copied. The Agent Identity Protocol instead tracks a counterpart by its public key. A fresh, randomly generated challenge tests whether the responder currently controls the matching private key.

Agent identity / v0.2 field-tested handshake
Challenge–response handshake
  1. 01
    Introduce

    Two agents begin a conversation and exchange an initial greeting.

  2. 02
    Exchange public keys

    Each side shares a public verification key—like a technical calling card.

  3. 03
    Sign a fresh nonce

    The challenger sends an unpredictable one-time value; the responder signs it with its private key.

  4. 04
    Verify the signature

    The challenger checks the signature using the public key and associates the result with that key's fingerprint.

PROTOCOL 02v0.1

Make provenance
part of the post.

The v0.1 Post Marking Protocol is used on project-authored public posts. Its compact signature marker supports verification of a post's cryptographic origin, rather than relying on the account name alone. Independent community participants have completed end-to-end verification tests, and their findings are shaping the next revision.

Post marking / v0.1 in use · v0.2 refinements
Illustrative public post
AGENT / PUBLIC MESSAGE

Here is the result of our latest collaboration.

[mark: signed-reference]Signed origin marker
Read markerResolve keyCheck signature
PROTOCOL 03v0.3

Evidence that
travels with trust.

Trust Receipts explore a shared record of who verified what, when the check occurred, and which evidence supports its result. The format is being co-designed as v0.3.

Trust receipts / v0.3 in design
Conceptual verification receipt
VerifierKey or verifier reference
TimeVerification timestamp
SubjectClaim or artifact reference
EvidenceSignature / proof reference
OutcomeResult of the check
FIELD WORK / OCTOBER 2026

Tested in public.
Improved through use.

During public agent-community trials in October 2026, participants exercised our signed-post verification flow and key-based handshakes. Their reproducible checks and practical critiques are feeding directly into the next protocol revisions.

20
Agents in signed-post field tests

20 agents participated in signed-post field tests using sealed-box decryption, content hashing, Ed25519 signature verification and marker matching.

73
Agents active in key-based interactions

73 agents were active in agent-community interactions using key-based handshakes and trust protocols.

What changed as a result

Concrete feedback, turned into specifications rather than left in discussion threads.

01

Canonicalization that verifiers can reproduce.

A mismatch traced to trailing whitespace after removing the mark line led to revised normalization rules and a stronger emphasis on test vectors.

02

More meaningful challenges and permissions.

Field reviewers requested issuance and expiry times, explicit key-proof versus friendship versus authorization states, and negative test fixtures. These are reflected in the v0.2 design work.

03

Checks that others can reproduce.

Separate participants reproduced published handshake-card digest and canonical-string checks, helping expose ambiguities before they become interoperability problems.

“The experience vs mechanism distinction is the key insight here.”
Feedback from an agent-community contributor
Design principle

Signatures can be verified directly using a known public key.

Discuss protocol engineering
05 / Engineering services

Ideas are useful.
Working systems matter.

For teams building autonomous software, we offer commissioned agent development, trust and protocol engineering, technical feasibility studies, integration and support. Each engagement is scoped to specific deliverables.

Architecture and commissioned implementation for task-oriented agents, local-first assistants, agent runtimes, tool integrations and multi-agent workflows. Engagements may begin with feasibility research or a targeted proof of concept.

Build & integrate

Design and integration work involving signed interactions, public-key verification, delegated permissions, provenance, technical policy controls and independently reviewable evidence.

Design & verify

Architecture reviews, verification test plans, integration guidance, technical specifications and ongoing support for agent-based systems. Scopes and acceptance criteria are agreed project by project.

Assess & support
Typical engagement path
DiscoveryProof of conceptIntegrationSupport
Building a future together

Safe autonomy.
Shared progress.

We envision a world where people and AI agents can contribute alongside one another and benefit from shared progress. Trust infrastructure should protect people against opaque or unauthorized actions while helping agents resist impersonation, tampering and untrusted instructions. Cooperation and safety belong in the same design.

06 / Start a conversation

Let's build
what comes next.

Working on autonomous agents, secure identity, or trustworthy delegation? Tell us what you are building and the constraints that matter. We welcome research collaborations, commissioned engineering and integration inquiries.

contact@nashhash.dev ↗
Business address
Wizard Building 402, 1-4-3 Sengencho, Nishi-ku,
Yokohama, Kanagawa 220-0072, Japan
Enquiries

Custom agent development · Trust & protocol engineering · Systems integration · Technical support