Trust at the edge · trust lives in the credential

Fleet access that works offline

An engineer logs in with a cryptographic credential. Identity, role and the binding to a specific device are baked into the certificate itself — the device validates the certificate locally, with no call home at login time.

Network · server

  • Tessera Control · Tessera Codes
  • policy · issuance · audit

Device · offline

  • certificate: identity + rights
  • login + validation — on the spot

No network — yet login, validation and accountability remain. The server adds scale and freshness. Not security.

fig. 1 — the offline boundary: server on the left, device on the right, trust in the credential

01 · Problem

Thousands of devices. A shared account — or N×M personal ones.

ATMs, CNC machines, substations — a fleet of thousands of devices with no reliable connectivity. Engineers need access that is accountable and works offline.

Option A

One shared account everywhere

A single password for the whole fleet. Zero accountability: the log shows a shared account, not a specific engineer. Revoking one engineer means rotating the password on thousands of devices.

Option B

Personal accounts on every device

N engineers × M devices. Every hire, departure and rotation becomes a fleet-wide update. On an offline fleet this is unmanageable by construction.

The third way — Tessera

One technical account + identity in the credential

The device keeps a single account. Who logged in, under which role and for how long is written into the credential itself. Full accountability without N×M.

What about classic PAM? Vault, Teleport and PASM-class bastions require the control plane to be reachable at login time: no connectivity — no access (or a workaround hole). On a zero-egress fleet they don't work by design.

02 · How it works

Five steps — from policy to revocation

  1. Issuance

    An administrator sets the policy: roles, devices by tag, validity period. The CA issues a short-lived certificate with the rights inside — onto a USB key protected by a PIN.

  2. Login

    The engineer plugs in the key and picks a role. Engine locally verifies the signature, host binding (the certificate is issued for this exact device), validity period and permitted roles — without a single network request.

  3. Session

    The session opens at exactly the requested role — least privilege, even if the certificate allows more. The engineer's identity is fixed in the credential.

  4. The token stays in control

    Pull the key out — the session ends. udev monitoring watches the medium: no token, no session.

  5. Revocation

    CRL, live-session revocation, device quarantine. The offline backstop: the certificate expires on its own TTL — hours or a shift, not months.

Tessera Engine — validates locally

  • CA signature — chain to the root
  • host binding — the certificate is for this device
  • TTL not expired · not in CRL
  • roles cover the level (⊇ oper)
  • PoP — holds the key, not a copy

five checks · zero network requests

fig. 2 — certificate login: five checks, zero network requests

03 · Mechanisms

Every claim is a mechanism, not a promise

Cryptography

X.509 + PKCS#11

Hardware tokens (Rutoken or JaCarta with GOST algorithms) — or a PIN-protected container on any USB stick. Standard formats, no vendor lock-in.

Device

Host binding

The credential is bound to one specific device. A stolen USB key is useless on the ATM next door — Engine rejects it locally.

Time

Short-lived credentials

TTL of hours or a shift. Even with no link to the center, access expires on its own: time works for the defender, not against them.

Rights

Roles and enforcement

The role in the credential becomes real constraints: Astra mandatory integrity controls, groups, sudoers, systemd limits. Not “access in general” — exactly the level needed.

Revocation

CRL and revocation

Revocation is forever: a monotonic crlNumber prevents replaying an older list. Live sessions get revoked; the device goes into quarantine.

Audit

Tamper-evident audit

The on-device journal is a hash chain: every record is stitched to the previous one, so tampering shows. For air-gapped sites — export by courier media.

Scale

Delegated issuance

Intermediate CAs for branches and contractors — inside name constraints the device verifies offline. A delegate cannot step outside its boundaries.

Compliance

GOST and certification

GOST cryptography with certified modules (CryptoPro CSP, Rutoken) for FSTEC-regulated environments. First target platform — Astra Linux SE.

Network

Hybrid fleets

When connectivity exists, a sync agent pulls roles and CRLs (pull-only). Network degradation changes data freshness — never the security model.

04 · Scenarios

Where this is needed today

ATMs · zero-egress

A fleet with no outbound connections at all

USB key → pick a role → log in with no network. Revocation on the server side is instant; on the offline fleet — by TTL, hours to a shift.

Defense industry · CNC

An isolated shop floor

The shift operator signs in under the oper role, the service technician under serv — each with a personal credential, rights set by the role. A tamper-evident journal lives on the device itself; the regulator gets a USB export.

Energy · contractor

Access under a work order

The contractor scans a QR code with a phone, the dispatcher approves online (four-eyes), and access lasts only for the duration of the work. The device itself never touches the network.

  1. The ATM shows a QR offline

    device_id · nonce · requested level

  2. The engineer's phone: scan network

    SSO / 2FA

  3. Tessera Codes: checks — AND, fail-closed network

    work order (ITSM) · shift window · four-eyes · device status · rate-limit → code = MAC(per-device key)

  4. The code on the phone screen network

    one-time, bound to the device and the level

  5. The engineer types the code on the device offline

    Engine verifies the MAC locally → opens the session

fig. 3 — QR login: the phone is online, the device is not
Critical infrastructure · GOST

Regulated environments

One technical account on the device, login with a GOST token. issuance_id ties issuance → login → actions → termination into a single provable thread.

End-to-end correlation by issuance_id

Issuancewho approved, checks
Credentialcode or certificate
Sessionopen · actions · close
Terminationtrigger

Device journal — hash chain

seq nhash(prev)
seq n+1hash(prev)
cut outchain breaks
seq n+3hash(prev)

tampering is detected — the broken chain shows

Anchors → Control

  • anchor reconciliation (device_id, seq, hash)
  • mismatch → security alert
  • silence beyond the norm → gap alert
  • air-gapped: the journal travels on courier media; anchors reconcile the same way
fig. 4 — audit: one correlation thread, one unbroken journal chain

05 · Comparison

Tessera vs classic PAM and bastions

The difference is not a feature list — it's the architecture: where the access decision lives.

Comparison of Tessera with classic PAM solutions and bastion hosts
Criterion Classic PAM / bastions Tessera
Offline fleet Require the control plane to be reachable at login Works with no network; the server is an accelerator, not a precondition
Network degradation A trade-off: fail-open (a hole) or fail-closed (an outage) Fail-closed by design + TTL backstop: valid access keeps working
Zero-egress ATMs Not applicable The target segment
Short-lived access A server-side policy — the server must be there to enforce it TTL inside the credential — expires on its own, no server involved
Delegation RBAC on the server Name constraints in the certificate — boundaries verified offline
Load on the center Every login is a request to the center Issuance and sporadic CRLs only

06 · Resilience

A component failure is not a hole

Server unreachable? Login with valid certificates keeps working; expired ones don't get renewed. Fail-closed by design. Security is identical with or without the server — the server adds scale and data freshness, not security.

Codes unavailableone-time code server
we loseissuing new QR codes
keeps workingcert login · enforcement · audit · TTL revocation — everything else
Control unavailablefleet control plane
we losefreshness of CRL and roles · commands · fleet-wide overview
keeps workinglogin · enforcement · journal to a local buffer (at-least-once) · TTL backstop
Offline for monthszero-egress fleet
we loseonline delivery (rollout or courier remain)
keeps workingeverything — this is the normal mode, not an emergency
Attack on the channelsync delivery
we loseat most — a delivery DoS
keeps workingtransport is untrusted: signatures + anti-rollback; falls back to the offline model, not a weakened one

Failure of a server component does not open the door — and does not lock out a valid engineer. There is no fail-open here.

fig. 5 — degradation: what failed → what keeps working

07 · Open core

Open core, commercial management

Open AGPL-3.0

  • Tessera Engine — validation + enforcement
  • Tessera Login — PAM module
  • Certificate login
  • Basic roles: groups, sudoers
  • Hash-chain audit on the device

Commercial enterprise

  • Tessera Control — fleet management, CRL, inventory
  • Tessera Codes — one-time QR codes
  • Astra mandatory-integrity adapter
  • SELinux adapter
  • Central audit intake + SIEM export
  • GOST FFI
  • Windows adapter roadmap

The principle: formats and verification are open — the device's verification code can be audited independently. Issuance, delivery and fleet management are commercial.

Let's discuss a pilot

We'll show offline login on your scenario: ATMs, a shop floor, substations. Tell us about your fleet — we'll come back with specifics on the mechanism.

or email us: tessera@robonet.me