Food & Beverage AI agent · Deforestation diligence
Deforestation Due Diligence AI Agent
Bind every consignment to the plots that produced it under Article 9(1)(d), score country risk against Implementing Regulation (EU) 2025/1093, and hold the statement for the person who lodges it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named submitter
- Plot-bound
- After launch
- Agent Care
What this agent does
Prepares the file, never the declaration
In
- 01
A consignment is planned, and Article 4(1) wants a statement lodged before placing or exporting.
- 02
A plot is supplied, and Article 9(1)(d) wants geolocation for every plot behind that commodity.
Reason
- 03
A plot sits in a named country, and Implementing Regulation (EU) 2025/1093 classifies its risk.
- 04
A consignment is scored, and Articles 10 and 11 want assessment first, then mitigation.
- 05
A consignment moves downstream, and Regulation (EU) 2025/2650 lifts the statement duty there.
Decide
- 06
A consignment leaves a micro or small holding, and 2025/2650 allows one simplified declaration.
- 07
A year closes, and Article 12(2) wants the system reviewed and the records held five years.
Out
- 08
A statement is lodged, and Article 6 lets a representative submit while the operator answers.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the file and tracks the Article 12(2) review. A named person at the operator, or an Article 6 authorised representative, lodges the statement — and the operator stays responsible either way.
Example workflow
One consignment, plots to lodging
- 1Consignment evidence receivedPurchase records, plot coordinates, supplier declarations or legality documents
- 2Consignment context assembledThe commodity, its heading, the country of production and the plots it was grown on
- 3Diligence file draftedThe plots, their coordinates, the gaps and completeness
- 4Controls appliedGeolocation-completeness checks, country-risk checks, date checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and nothing is lodged at any of them — the agent is assembling, and the sourcing lane opens at the completeness gate.
Goes to the operator or its authorised representative to lodge.
Adds a sourcing analyst read first.
Sourcing review
The file is held with its consignment, its plots and the gaps between them.
Lodge · Append evidence · Send to sourcing review
Lodged — under Article 4(1)
- 6Supplier and plot records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedGeolocation coverage, risk scoring, analyst corrections and what the review found
Corrections
Each sourcing correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Lodging the statement under Article 4(1).
- Concluding that risk is no more than negligible.
- Reviewing the diligence system, Article 12(2).
- Appointing a representative under Article 6.
Automation boundary
Agent acts unaided
- ✓Bind each consignment to the plots that produced its commodity.
- ✓Check that Article 9(1)(d) geolocation is complete for every plot.
- ✓Score country risk against the published benchmarking.
- ✓Track the Article 12(2) review and the record retention.
Nothing is lodged or declared except by a named person, inside the agreed boundaries.
- Judging whether production was legal in its country.
- Telling a competent authority the file is complete.
- Deciding the risk class a country is worked at.
- Changes to supplier, plot or legality records.
Example output
One consignment, annotated
Application opens 30 December 2026 for large and medium operators and 30 June 2027 for micro and small enterprises; this record is what one consignment carries beforehand.
Diligence file · single consignment
Illustrative example- Consignment
- Cocoa butter inbound, Article 9(1)(d)
- Recorded as
- Plot given as a village name, no coordinates
- Commodity
- Cocoa, derived
- Evidence of record
- Supplier declaration, 3 August 2026
- Confidence
- Held unlodged
- Held for
- The operator or its authorised representative
As received
Read off the supplier declaration and the purchase record, and no further than those two reach.
What the record holds
- Supplier declaration
- Plot coordinates
- Legality document
Why no lodging here
Whether the statement may be lodged is an Article 4(1) act for a person.
What the score decides
Below the configured threshold the file picks up a sourcing read before it is lodged.
Action
LodgeAppend evidenceSend to sourcing reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every consignment
From the plots behind it
Where the evidence is used
Our supply-chain traceability agent answers where a lot went; this page answers whether that lot may be placed on the market at all. The forced-labour import evidence agent is UFLPA at the US border about labour; this is a positive declaration before the EU market about land — different market, different regulator, different evidence.
A plot, and who says so
Article 9(1)(d) wants coordinates for every plot, and Article 33 puts the statement in the Information System launched 4 December 2024 and reopened in June 2026.
What was checked, and not found
The Annex II declaration wording could not be retrieved verbatim, and it is not established that Annex II carries a signature or name-and-function block; EUR-Lex renderings truncated ahead of the annexes on every attempt, and only Commission pages confirm the substance — no or only negligible risk. This page rests on Article 4(1) and the Article 12(2) review instead.
Build an evidence trail
The consignment, the plot it came off and the person who lodged it stay on the statement.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Purchase and supplier records
- ERP
- supplier portal
- Consignments and commodities
Plot and geolocation data
- GIS files
- coordinates
- Plot polygons and points
Legality and certification
- Permits
- scheme records
- Country-of-production evidence
Agent
Deforestation diligence file
- Reads the plots
- Assembles the evidence
- Holds for the operator
Risk and benchmarking
- Country classification
- Article 12(2) review dates
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six tests between the model and the lodging
Six tests, and a consignment must pass every one of them. What comes through is charted in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a file may stand
- L3
- Puts the lodging in a person's hands
- L4 – L5
- Keep the consignment and the plot behind it
- L6
- Holds the statement unlodged when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the diligence file that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the file an authority reads
- E1Final-output evaluationDid the file record what each consignment actually carried?
- E2Step-level evaluationDid the agent read the right plot, the right country and the classification in force?
- E3Tool evaluationDid it read and write the correct consignment record and the correct plot?
- E4Confidence calibrationDo low-confidence files actually attract more sourcing corrections?
- E5Slice evaluationHow does performance change across specific commodities?
- E6Business outcomeHow many files needed a correction before the operator lodged?
Floor — the file the operator answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed at the stage where it first shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- JP-03
Stale plot record read
The coordinates read are not the ones now supplied.
Stage gathers
The plots, the country, the risk and the dates
02 · Reasoning
2 modes
- JP-04
Plot asserted, not shown
A plot is called covered without its coordinates.
- JP-06
Draft act read as adopted
The proposed product scope is worked as binding.
Stage proposes
The plots, their risk scores and completeness
03 · Tool / write
2 modes
- JP-02
Thin file passed forward
A file moves on without the sourcing read.
- JP-05
Evidence bound to wrong plot
A record is filed against the wrong plot.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- JP-01
Lodged, evidence unrecorded
The statement shows a lodging but not what supported it.
Stage returns
The file an operator lodges and an authority reads
05 · Change / Version
1 mode
- JP-07
Silent scope regression
A configuration change moves the scope, not the file.
Stage tracks
Model, prompt, sourcing rules and evidence fields
- Sev-1 · a statement lodged on no plots
- Sev-2 · wrong evidence reaches the file
- Sev-3 · source degrades, file holds unlodged
Affected slices
Cocoa absorbs the corrections
A consignment-level plot-provenance figure can read clean while cocoa carries most of the rework. Nestack reports the correction rate by commodity, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Cocoa butter, paste and powder | 8.7% | 3.6× | Review | |
| Green and roasted coffee | 6.2% | 2.6× | Review | |
| Palm oil and its fractions | 3.9% | 1.6× | Watch | |
| Wood packaging and paper | 1.4% | 0.6× | Normal |
Bar: correction-rate lift vs. wood-packaging baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a missing polygon costs
A cycle ends when the plotless consignment is a standing case. That suite is what the next statement assembled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on cocoa consignments.
- 02
Diagnose
The plot that arrived as a village name where the rule wanted coordinates is read back until one cause remains.
- 03
Improve
Changes leave numbered, and the consignments that forced them travel attached.
- 04
Verify
One consignment case still failing is enough to stop the whole release.
- 05
Learn
The case is kept, and the sourcing rules are rewritten alongside it.
Learn → Detect
The return edge. The next file is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, evidence assembly, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Commodity-scope and automation-boundary discovery.Week 1
- 02Plot, supplier and legality sources.Week 1, Week 2
- 03Consignment-to-plot and country-risk-class mapping.Week 1, Week 2
- 04Plot and geolocation ingestion.Week 2
- 05Consignment, plot and record binding.Week 2, Week 3
- 06Completeness scoring and review routing.Week 3
- 07Operator lodging workflow.Week 3, Week 4
- 08Supplier and plot-system integration.Week 5
- 09Geolocation and risk cases.Week 4, Week 5
- 10Guardrails and lodging controls.Week 4
- 11Consignment-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne commodity, one country | ProductionProduction diligence workflow | AdvancedMultiple commodities / countries |
|---|---|---|---|
| Introduced at Pilot | |||
| Evidence assembly to your consignments | ✓ | ✓ | ✓ |
| Operator or representative release | ✓ | ✓ | ✓ |
| Plot-provenance baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by commodity | — | ✓ | ✓ |
| Lodging workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Supplier-record integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-country sourcing | — | — | ✓ |
| Cross-commodity evidence packs | — | — | ✓ |
| Large plot registers | — | — | ✓ |
| Multi-country risk controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, consignment volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your commodities in scope and the countries behind them
→Commodity scoping and consignment capture
Week 102Representative plot, supplier and legality records
→Record binding, risk logic and the evidence baseline
Week 203Your geolocation data under Article 9(1)(d)
→Plot mapping, consignment binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Plot, supplier and legality-source assessment, then integration setup
Week 205Statements you would not want checked
→Risk cases and the evaluation suite
Week 406What no diligence statement may guarantee
→Completeness scoring, review routing, guardrails and release controls
Week 307A named person at the operator to lodge
→Lodging workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Each bar is sized to the weeks that phase really needs, and the fifth of them therefore doubles up.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Diligence workflow discovery, plot mapping and the automation boundary
W2Source integration and the plot-provenance baseline
W3Evidence assembly, risk logic and release controls
W4Evaluation suite, geolocation cases and failure-mode testing
W5Record integration, pilot consignments and targeted corrections
W6One sourcing season run under the authorised representative, then Agent Care handover
- Reading the band
- Every band runs only across the weeks its own work is named for. The fifth takes on two because the work does.
- At the end of W6
- When the plot record validates, Agent Care picks the agent up.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Food & Beverage AI agent
Build a deforestation diligence agent ahead of 30 December 2026.
Show us one consignment and the plots behind it. Not a lot history. A positive declaration before the EU market, under Article 4(1), that the land was never cleared. Large and medium operators apply from 30 December 2026, four months out. Traceability is a different agent.
- Typical delivery
- 4–6 weeks
- Deployment
- Your consignments
- Named submitter
- Plot-bound
- After launch
- Agent Care available
Nestack Agents · Deforestation diligence
AGT-FB-16 · Agent Care available after launch