Early access for Channel Partners and ISVs opening soon. Learn more →

Edge program · In certification

FastYoke at the edge — integrity you can attest

The same single Rust binary and per-tenant SQLite that run the cloud platform, running on edge hardware. Boot integrity is measured, attested, and surfaced as a compliance control — an edge fleet you can prove is trustworthy, not just claim is.

The stack sets the envelope

FastYoke's runtime is a Rust/Axum binary, per-tenant SQLite (our "Fleet of Files"), a QuickJS-in-wasmtime sandbox, pure-Rust tract ONNX inference (CPU, no GPU required), and LiteFS/Litestream sync — shipped as a Debian-bookworm (glibc) container. That is the whole dependency footprint, and it is what defines the hardware envelope below: modest RAM, standard storage, no accelerator, no exotic OS.

If a device runs a mainstream Linux container engine and clears the floors below, FastYoke runs on it.

Related reading: Running your code — safely, at native speed: why FastYoke chose WebAssembly — why the QuickJS-in-wasmtime sandbox is part of that footprint.

Two ways to run at the edge

Edge Agent or Edge Node

The same engine, sized to the job — a thin client that bridges local I/O and buffers offline, or a full local-first runtime.

Edge Agent — thin client

Connectors, snapshot polling, and local-I/O bridging. Buffers while offline and syncs to the cloud over HTTPS. Heavy compute stays in the cloud. Floor 1 GB RAM / 8 GB eMMC (2 GB / 16 GB preferred). Typically a cellular router with a container engine or a small SBC.

Edge Node — full runtime

The full binary on device: per-tenant SQLite, wasmtime/QuickJS scripting, tract ONNX, LiteFS local-first sync, and a local API/UI. Floor 2 GB RAM / 32 GB NVMe or eMMC (4 GB / 64 GB+ preferred; no SD cards in production). Typically an x86-64 mini-PC or an industrial DIN-rail gateway.

Architecture policy

x86-64 first, ARM64 first-class

x86-64 (amd64) is the primary target — the container ships and is certified here first. ARM64 (aarch64) is a first-class second target via the multi-arch image. Other instruction sets (RISC-V, 32-bit ARM) are out of scope for now.

The differentiator

Boot integrity you can attest

Generic gateway vendors ship hardware. FastYoke ships an edge fleet whose boot integrity is continuously attested and auditable — down to the PCR.

Hardware-attested integrity is a hard requirement for the Certified tier. A device qualifies when it provides:

  1. TPM 2.0 — discrete preferred; firmware/fTPM acceptable on x86-64.
  2. UEFI Secure Boot — or an equivalent verified boot chain on ARM64.
  3. Measured boot — PCR extension that FastYoke reads at agent startup.

FastYoke reads the TPM quote and measured-boot PCRs on start and surfaces edge-node integrity attestation as a compliance control in your Trust Center and auditor room. Your SOC 2 / compliance story extends all the way down to the edge — not "the gateway is locked in a cabinet," but "here is the measured, signed proof that this node booted the software we shipped."

Certification levels

What each tier means

LevelBarUse
CertifiedPasses the conformance suite, meets the attestation requirements, and the vendor offers OTA/fleet managementListed and supported
SupportedPasses the conformance suite; attestation is partial or via an add-on moduleRuns in production, best-effort
CommunityBoots and runs; no attestation or management guaranteesPilots and evaluation only

Device matrix

Hardware we are certifying

Representative devices across both architectures and both form factors. Status reflects where each device is in our certification program today.

DeviceArchClassAttestationStatus
OnLogic Karbon/Helix, NUC-class fanlessx86-64Edge NodefTPM/dTPM + Secure BootCertified reference (target)
Dell Edge Gateway 5200 / Lenovo ThinkEdge SE30–SE50x86-64Edge NodeTPM 2.0 + Secure BootIn certification
Advantech UNO / EISx86-64Edge NodeTPM 2.0In certification (rugged)
Moxa UC-seriesx86 / ARMEdge NodeTPM (model-dependent)In certification
Digi IX / CradlepointARM / x86Edge AgentSecure boot + secure elementIn certification (cellular fleet)
Teltonika RUTXARMEdge AgentLimitedSupported (target)
Raspberry Pi CM4/CM5 + TPM moduleARM64Agent or NodeTPM via add-on moduleSupported when fitted with a TPM module
NVIDIA Jetson OrinARM64Edge NodeTPM (carrier-dependent)Community

Conformance

What certification tests

Every Certified device passes the same suite. Here is exactly what that means.

Builds and boots

The multi-arch image builds and boots on the target (amd64 today; arm64 is the second target). Secure Boot is enrolled and an unsigned or tampered image is rejected.

Runs under the floor

Edge Node: wasmtime scripting, tract ONNX, and per-tenant SQLite all pass under the RAM floor. Edge Agent: the offline buffer and resume-sync survive a 24-hour disconnect with no data loss.

Syncs and reconciles

Litestream/LiteFS round-trips to the cloud and reconciles cleanly after reconnection.

Attests its boot

The TPM 2.0 quote and measured-boot PCRs are read and surfaced as a Trust Center control. A remote FastYoke update applies cleanly over the vendor's OTA/fleet path.

Speaks the local bus

Local I/O is enumerated for the target vertical — RS-232/485, CAN, Modbus, or GPIO as applicable.

Survives the soak

72 hours at the rated temperature under representative load, with no thermal throttling or data loss.

Where Edge earns its keep

The regulated imaging fleet

Preclinical irradiator platforms, benchtop small-animal imagers, dental research systems, veterinary radiography — imaging devices that live in shielded rooms, produce the highest-scrutiny session records in the organisation, and cannot wait for a cloud round-trip during a live exposure.

Every exposure is a regulated event. The device console captures the delivered dose, the kVp, the filtration, the geometry, the operator identity, and the subject cohort — and every one of those fields is required by the study protocol, the animal-care protocol, and the retention regime that follows. Miss the timestamp and take a finding at the animal-care review. Miss the operator signature and take an e-records deviation at the sponsor audit. Miss the dose-rate calibration link, and the paper the study feeds cannot be published.

FastYoke Edge sits between the device and the compliance regime. An Edge Agent talks to the console over RS-232/RS-485, a DICOM MPPS feed, or a vendor-proprietary file drop — pulls the session record the moment the exposure ends, signs it with the tenant's key, writes it to the on-device per-tenant SQLite, and syncs to the tenant's cloud tier as soon as network returns. The device console is unchanged. The operator's workflow is unchanged. The record that lands in the research-data-management system carries the boot-attestation evidence for the device that produced it — so an auditor doesn't have to trust the record, they can verify that the device that wrote it was in a known-good state at exposure time.

The mechanism is compatible with a Part 11 retention regime — the append-only event log, per-tenant crypto, and RBAC primitives that Part 11 asks for are already how FastYoke stores every other tenant record. Mapping the specific controls to a specific regulatory posture is a conversation, not a checkbox.

Named use cases

Three shapes of regulated imaging

Different buses, different retention windows, same session-record posture.

Preclinical irradiator platforms

Research suites running 30–150 exposures/day across mouse and rat cohorts. Session record includes cohort ID, dose (Gy), dose rate (Gy/min), kVp, filtration, geometry, and operator badge. Retention runs to the study lifetime plus regulatory tail — typically 10+ years. The Edge Agent bridges the console over RS-232 or a DICOM MPPS feed; the record signs and syncs before the operator has removed the cassette.

Benchtop small-animal imagers

Small-animal cone-beam CT, PET/CT hybrids, dental research systems. Same session-record posture, different bus — typically a DICOM SR stream or a vendor-proprietary CSV drop into a shared folder. The Edge Node absorbs the drop, extracts the structured fields, cross-references the subject and operator, and produces the same signed session envelope as the irradiator path.

Veterinary radiography

Working veterinary clinics running research-grade imaging (equine, exotic, zoo). Owner consent, subject welfare, and case-record chain-of-custody sit alongside the imaging data. The Edge Node keeps everything local for the referring vet and syncs to the specialty group's cloud tier when connectivity is available — the horse trailer at a rural clinic doesn't have LTE.

::

::

Where Edge earns its keep

The cold-chain and yard fleet

Refrigerated trucks and trailers, DC gates and dock scales, last-mile routes — logistics operations where the record has to land before the network catches up, and where the connectivity story is a national-scale honest problem.

Every mile is a regulated event. A refrigerated trailer running a mixed produce load carries a temperature-record obligation the receiver will absolutely audit at the dock — miss a five-minute excursion and the receiver either accepts the load with a claim attached or refuses it outright. The reefer unit already produces the reading on its own Modbus bus. What it doesn't do is timestamp it against dispatch, GPS-anchor it, sign it, and survive the 400-mile stretch of I-40 through the Texas panhandle where LTE is a suggestion.

FastYoke Edge sits in the cab. An Edge Agent talks to the reefer over RS-485 or J1939, pulls the temperature reading on the duty-cycle interval the load called for, tags it with the load number and the driver, signs it with the carrier's tenant key, buffers to on-device SQLite, and syncs to dispatch when the truck rolls into a working tower. The receiver sees the signed record before the trailer is dropped. The chain-of-custody story runs from origin to delivery with cryptographic receipts at every hop — compatible with an FSMA Sanitary Transportation retention posture, without asking the driver to touch a tablet.

The same Edge Node runs at the DC gate. A camera reads the container number, the gate boom opens against the appointment window, and the yard system records the arrival — locally, sub-second, then syncs. A three-second cloud round-trip to open a gate is not a technical inconvenience; it is a queue backing up onto the highway.

Named use cases

Three shapes of edge logistics

Different buses, different retention windows, same signed-record posture.

Cold-chain telematics

Reefer trailers running temperature-critical loads (produce, seafood, pharma, biologics). Edge Agent bridges the reefer control unit over Modbus/J1939, timestamps against dispatch and GPS, signs each reading with the carrier's tenant key, and buffers through no-signal corridors. The receiver's audit runs against a signed record set, not a printed thermograph strip.

DC gate and yard management

Terminal gate cameras, dock-door scales, container-number OCR. The Edge Node correlates arrivals against the appointment book, opens the boom, weighs the pallet against the outbound BOL, prints the receiving label — all local. Cloud sync is for reporting and reconciliation, not for opening the gate.

Last-mile route replanning

Driver tablets running the routing engine on device. When dispatch adds a stop, the algorithm reruns locally — the driver in the signal-dead cul-de-sac gets the new sequence without waiting for a cloud round-trip. Delivery proof (signature, photo, geofence) signs and syncs as soon as bars return.

::

::

Where Edge earns its keep

The connected store

Point of sale, receiving dock, backroom count — retail operations where a network outage is a business emergency, not a technical inconvenience, and where "the ISP is down" is not an acceptable answer for a customer holding a card.

Every transaction is a customer relationship. When the store's ISP hiccups mid-swipe, three things need to keep working: the card reader has to take the payment on the local acquirer path; the cash drawer has to open on the receipt-printer command; the inventory has to decrement so the next customer isn't sold the last unit twice. Cloud-only POS gives you an outage window the size of the ISP's next fibre-cut. Local-first POS gives you a store that runs the same on a bad-network day as on a good one.

FastYoke Edge sits behind the counter. The Edge Node runs the POS locally — card reader over USB/HID, receipt printer over serial, cash drawer over the drawer-kick line, scale over RS-232 — writes every sale to per-tenant SQLite, signs the transaction, and store-and-forwards to the corporate cloud tier when the pipe comes back. The receipt has the customer's name on it. The close-of-day report has every sale on it. The inventory decrement survived the outage.

The same Edge Node runs receiving at the back door. A scanner reads the pallet label, the ASN comes down from corporate, the Edge Node reconciles line-by-line and flags the discrepancy before the truck leaves. Receiving-window claims that would have been argued from a paper log are argued from a signed record.

Named use cases

Three shapes of edge retail

Different peripherals, different retention windows, same offline-first posture.

Local-first POS

Card reader, receipt printer, cash drawer, kitchen printer, backroom SKU lookup. The Edge Node runs the sale locally, prints the receipt, updates inventory, and store-and-forwards to the corporate cloud tier. An ISP outage is a routing problem, not a revenue problem.

Receiving and ASN reconciliation

Handheld scanners at the back door talk to the Edge Node, which reconciles arrivals against the Advance Shipping Notice, flags overages/shortages/damages, and signs a receiving record before the vendor's truck pulls out. Discrepancy claims run from signed data, not memory.

Loss-prevention analytics on device

tract-ONNX runs the LP model locally against the store's camera streams. Only tagged incidents leave the store — for privacy, for bandwidth, and for the latency budget that catches an incident before the actor leaves the aisle. Video stays behind the counter unless there is a reason to ship it.

::

::

Where Edge earns its keep

The clinic front desk and patient throughput

Ambulatory check-in, patient throughput boards, facility access and audit — the administrative surface of a clinic or hospital operation, where the record has to land locally when the ISP drops and the compliance surface is HIPAA administrative safeguards.

Every visit is a coordinated handoff. When a patient arrives at an ambulatory clinic, the check-in kiosk verifies identity, scans the insurance card, settles the copay, and marks the appointment as arrived — four systems talking to each other on a network that fails at exactly the wrong moment. Cloud-only front desks either turn patients away or fall back to paper that gets keyed in later, at which point the arrival time is a guess and the copay reconciliation is a spreadsheet. Local-first front desks keep the schedule running through the ISP outage and flush to the corporate cloud tier when the pipe returns.

FastYoke Edge sits at the front desk and on the ward. An Edge Node runs the check-in flow locally — kiosk, insurance card scan, driver-license verify, copay POS, throughput board updates — and store-and-forwards to the corporate cloud tier when the pipe comes back. Facility badge readers and access-controlled room events (the medication room, the medical-records room, the IT closet) land on the same signed audit trail. The Patient Flow Yoke's throughput boards render off the local SQLite so a census view updates in real time regardless of the corporate network's mood.

The scope of the vertical is deliberate. FastYoke Edge does not sit in the clinical decision path — no medication administration verification, no vitals interpretation, no point-of-care result alerting. Those surfaces are FDA-regulated medical device territory and require a separate compliance program the platform does not carry. The Edge Node handles the administrative workflow — appointments, arrivals, throughput, access, audit — and hands off to a certified clinical system for anything that mediates a treatment decision.

The mechanism is compatible with a HIPAA administrative-safeguards regime — signed access records, per-tenant crypto, and the append-only event log that HIPAA §164.312(b) asks for are already how FastYoke stores every other tenant record. As with the preclinical imaging story, mapping specific controls to a specific regulatory posture is a conversation, not a checkbox. Tenants handling PHI pair the deployment with the BAA / HIPAA posture add-on.

Named use cases

Three shapes of edge clinical administration

Different peripherals, different retention windows, same offline-first workflow posture. Administrative surfaces only — the platform does not sit in the clinical decision path.

Ambulatory clinic front desk

Check-in kiosk, insurance card scan, driver-license verify, copay POS. The Edge Node runs the front desk locally so the clinic keeps taking patients when the cable modem drops — check-ins queue, copays settle to the local acquirer, insurance verifications flush when the pipe comes back. Administrative workflow only; no clinical data path.

Patient throughput and census

The Patient Flow Yoke's throughput boards run against the Edge Node's local per-tenant SQLite. Room-turn events, bed-status updates, and census views render in real time regardless of corporate-network state. A ward board that stays live through a Wi-Fi outage is worth more than one that goes blank at the wrong shift change.

Facility access + audit trail

Badge readers at the medication room, medical-records room, and IT closet land on the same signed append-only audit trail as every other tenant event. HIPAA §164.312(b) audit controls without a separate access-management SaaS. Access decisions run locally; the audit log syncs to the corporate cloud tier.

::

::

Where Edge earns its keep

The service bay and the maintenance depot

Shop-floor diagnostics, alignment rigs, fleet fuel islands, emissions test lanes — automotive operations where a warranty claim, a regulatory audit, and the driver stranded at the fuel pump are three sides of the same "the record has to land locally first" problem.

Every repair order is a warranty case in waiting. When a diesel truck rolls into the bay with an emissions fault, the shop's scan tool pulls the DTC and the freeze-frame data from the ECM over J1939 or OBD-II. That capture is the warranty evidence. Six months later, when the manufacturer denies the claim, the argument runs on whether the scan tool output was signed, timestamped, and bound to the correct VIN — not on whether the technician remembered which port they connected to. The scan tool already produced the record. What it doesn't do is timestamp it against the RO, sign it with the shop's tenant key, and survive the shop-management-system provider's next monthly outage.

FastYoke Edge sits in the shop. An Edge Node ingests the scan tool's output over USB or the bay's diagnostic Ethernet, binds it to the open RO, correlates it with the ECU photo the technician just snapped, signs the whole envelope, and pushes to the shop management system when the network is present. The manufacturer's warranty audit runs against a signed record set, not against a photocopy in a file folder.

The same Edge Node runs the fuel island at the fleet maintenance depot. Fuel dispenser, tank-monitor console, driver-badge auth, odometer capture — all local. When the depot's internet is down, drivers still fuel, and the reconciliation lands the moment the pipe comes back. Emissions test lanes get the same treatment, with signed VIN + dyno + tailpipe records that satisfy the state DMV's chain-of-custody requirement.

Named use cases

Three shapes of edge automotive

Different buses, different retention windows, same signed-record posture.

Shop-floor diagnostics

Scan tools, code readers, alignment rigs, brake dynos — each with its own USB, serial, or diagnostic-Ethernet output. The Edge Node ingests, binds to the open RO and VIN, correlates with technician photos and time-in-bay, and signs the warranty-evidence envelope before it ever leaves the shop.

Fleet fuel island + telematics ingest

Fuel dispensers, tank-monitor consoles (VeederRoot and equivalents), driver-badge auth at the pump. The Edge Node keeps drivers fuelling through a depot-internet outage; telematics from the vehicles pre-loads the maintenance schedule as trucks roll onto the lot. Reconciliation to the fuel-card provider flushes when the pipe returns.

Emissions + environmental compliance

Emissions test lanes, paint-booth VOC monitoring, cure-cycle timing, air-quality sensors. Signed, timestamped, tamper-evident chain-of-custody for state DMV and EPA-facing audit surfaces. Compatible with a Clean Air Act retention posture — the mapping is a conversation.

::

::

Why on-device across the fleet

The environment breaks connectivity by design. Lead shielding in an imaging suite. Masonry walls on a hospital ward. Metal roofs and dense racking in a DC. Signal-dead corridors on the interstate. A crowded storefront during a Friday-night surge. A depot's cable modem on the day its provider is migrating equipment. A cloud-only capture path either drops the record or produces "it was missing when the auditor asked" — the exact failure mode Edge is built to eliminate.

The bus produces the record; the network sees it after. The imaging console, the reefer unit, the card reader — they have already produced the artifact before any network stack touches it. Local-first capture with tenant-key signing makes that record durable, tamper-evident, and portable before the Litestream/LiteFS sync completes.

Retention runs long. Study data on a mouse cohort in 2026 may need to be produced for a sponsor audit in 2036. FSMA transportation records run multi-year. Retail receipt archives sit in tax-audit territory. Per-tenant SQLite plus optional Sovereign Git Vault materialisation gives you a retention path with cryptographic receipts — not "trust the vendor is still in business in 10 years."

The competitive edge is in the metadata. Protocol geometry and filtration recipes. Routing algorithms and dispatch heuristics. Loss-prevention models and pricing curves. Data-sovereignty concerns push against cloud-only capture even when the connectivity story would support it.

Third-party product and company names are the property of their respective owners. Inclusion in this list indicates a certification target or compatibility intent and does not imply a completed certification, partnership, or endorsement.