Trust Carbon and HexStellar bring more efficient AI to offline fieldwork

A field team should not have to wait for a network connection to record what it sees. Trust Carbon is building its environmental evidence workflow around that constraint, while its collaboration with HexStellar tests whether more efficient computation can make the app more useful on the devices people already carry.

For the broader market context, see our 2027 carbon market outlook on verifiable offline MRV and field evidence.

The collaboration also includes TrustBio, which focuses on biodiversity and forest inventory. A TrustBio newsroom report published in September 2026 said the technology was already working in apps and being tested in field and real-world settings, attributing that account to founder Brayon Pieske. [1] This is a report of integration and testing, not a published device-by-device benchmark.

The work starts with a person at a project site

Consider a trained collector visiting a forest plot with a project assignment already on the phone. The example is a workflow illustration, not a report from a named project. Before leaving coverage, the team downloads the relevant project-area map and assignment. In the field, the app can present the next methodology step, indicate the location to visit and keep the observations connected to the task. The collector records measurements, photographs, position and context while the device is offline.

Pre-downloading matters. An offline app cannot retrieve a missing map tile or a newly assigned task from a server with no signal. Trust Carbon's workflow is designed to keep downloaded maps, assignments and step status on the device, then preserve field records locally until a connection is available. Its public platform describes offline collection and automatic synchronization after connectivity returns. [2] A queued upload may still need a valid session, adequate connection and successful server response; a record should be treated as received only after the app confirms it.

The on-device checks are part of that same sequence. Location and device-integrity indicators can flag conditions that need review, such as a suspicious location source or a compromised device. They do not make a false observation impossible or establish that a species, land right or carbon quantity is correct. A legitimate collector may also encounter weak GPS under canopy. The useful result is a visible exception that a supervisor can examine, not an invisible assertion that every record is trustworthy.

When connectivity returns, the app queues the locally saved material for automatic synchronization. Authorized project developers can then review the uploaded records, assignments, maps and exceptions in the web dashboard, subject to project permissions. The evidence stays associated with the task that produced it. That continuity helps a reviewer ask specific questions: Which step was performed? Which photograph and coordinate belong to it? What was flagged on the device, and what still needs a human decision?

What HexStellar contributes, and what remains unmeasured

HexStellar provides computational acceleration and efficiency technology. Its public Enterprise Runtime description presents a licensed path for compatible workloads in applications and other computing environments. Its published quantitative campaigns cover defined server workloads, including AI inference on one NVIDIA H100; those numbers describe that hardware and those test conditions. They do not establish a speed gain or battery-life figure for a Trust Carbon phone. [3]

Brayon gave the newsroom this account of the current app work:

That statement conveys the founder's observation, not an independently reproduced measurement. “Greater autonomy” refers to the possibility of longer operation between charges, not to a published number of extra hours. The claim should be tested separately for each supported device and task. A useful comparison would run the same input on the same phone and app version with the relevant computation enabled and disabled, then report response time, energy or battery use, temperature and whether the output changed. Tests should also check behavior without a network and whether the original field record remains intact through later synchronization.

That distinction is especially important because offline collection and offline AI are different capabilities. A collector can capture and store a record without a connection even when a particular AI feature is unavailable locally. TrustBio publicly describes local inference in specific components, such as bioacoustics and trunk analysis. This does not mean every Trust Carbon AI function runs on every phone without a network. [1]

Three products, distinct responsibilities

Trust Carbon organizes carbon-project evidence for measurement, reporting and verification (MRV). TrustBio organizes biodiversity observations and forest-inventory records, with its own biological and forestry workflows. HexStellar focuses on computational work. Their collaboration connects related problems: data must be gathered in the field, interpreted in context and made reviewable without requiring a permanent connection. It does not make the three products interchangeable. [1]

A species suggestion is an aid to a qualified field team, not a confirmed identification by itself. A forest measurement may inform a project model, but it does not become an issued carbon credit. TrustBio's species and inventory information can provide environmental context beside carbon observations where a project integrates and authorizes that data; a record in one product is not automatically imported into every project in the other.

For the person doing the collection, the gain is a clearer assignment and a durable record of what was actually observed. For a project developer, it is the ability to review incoming evidence and send a targeted request when a step is incomplete. For an authorized investor or project owner, a permissioned view can show progress and supporting material. None of these views guarantees a financial return or replaces the review obligations of a carbon standard.

A local team needs more than a faster app

Offline maps and guided steps can make defined tasks more accessible to local farmers, community members and field workers who know the terrain. That is a practical reason to invest in lower device consumption: a phone that stays useful through a long field day is easier to include in a work plan. Yet usability must be tested with the people and devices involved. Training, supervision, appropriate languages, consent, data access and fair compensation remain project responsibilities.

Good task design also preserves the limits of expertise. A guided method can prompt for a photograph or diameter measurement; it cannot decide whether a site's sampling plan is scientifically sufficient, whether land rights are settled or whether a community has agreed to the use of its data. Those decisions belong to the responsible professionals and participants. More efficient computation has value only if it helps them see and correct uncertainty.

Evidence infrastructure is not a carbon-credit issuer

Trust Carbon supplies digital infrastructure for collecting, organizing and sharing environmental evidence. It does not sell or issue carbon credits, operate as a registry or replace an independent third-party audit. Its app can surface integrity indicators and a traceable collection history; it cannot certify its own climate outcome. Under a crediting program such as Verra's Verified Carbon Standard, the project proponent, independent validation/verification body and program each retain their roles. [4]

The right test of this collaboration is therefore practical and inspectable. Can a trained team complete the assigned field steps without a network? Do the records sync with their context intact? Can an authorized reviewer distinguish the original observation from an AI suggestion, a device warning and a later correction? Does a measured efficiency gain on a particular device preserve the same output? Those answers, when published with the workload and measurement conditions, will tell project teams much more than a universal promise of speed or fraud prevention.

For now, the collaboration's reported app and field tests mark a useful direction: bringing computation closer to the person who first sees the evidence. The next milestone is a public, comparable account of what that changes for a real field task, on a specified device, without weakening the chain of review.

TRUST CARBON NEWSROOM

Beatriz Siqueira · Product reporting. Factual corrections: [email protected].