Nutrition-Guidance AI Agent

What this agent does

Gives information, not a prescription

Ingesting the configured nutrition library, the club's referral list and whatever the member has logged.

Normalising each logged entry against the food database, and carrying it forward with its source.

Reason

Screening each question for eating-disorder, clinical and doping topics first.

Drafting general information, meal ideas and macro ranges from the approved library.

Checking the member's configured state against the dietetics licensure rule set held for that state.

Decide

Suppressing out-of-scope topics rather than answering them more carefully.

Holding anything individual or product-specific for the named dietitian.

Out

Retaining the suggestion, the profile it was drawn against, the referral offered and the edits made.

Execute write actions only inside the approval boundaries agreed during implementation.

→ Product statement

The agent supplies general information; the named dietitian decides anything individual, and the operator answers for what is said.

Example workflow

One question, intake to referral

AgentHuman

1Member question receivedMember app, kiosk, coach hand-off or a logged food entry

2Context assembledLogged entries, the configured state, flags on the member record and the referral list

3Suggestion draftedGeneral information, meal ideas, scope notes and confidence

4Controls appliedTopic screening, scope checks, state licensure rules, supplement suppression and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing individual leaves at any of them — the agent answers generally, and the dietitian's lane opens at the gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the club dietitian to release.

Low confidence

Adds a scope review first.

Dietitian review

The suggestion is held with the profile it was drawn against and the confidence.

Release · Edit · Refer to a professional

Released — sent to the member▼

6Member app updatedOnly where write access and approval policy allow it

7Outcome evaluatedEdit distance, referral outcomes, suppression hits and post-release corrections

Edits

Every dietitian edit is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Medical nutrition therapy for a disease or condition.

Recommending a supplement product to a tested athlete.

Engaging a member flagged for disordered-eating risk.

Setting an individual calorie target or rate of loss.

Automation boundaryAgent acts unaided

✓Answer general, population-level nutrition questions.

✓Suggest meals and recipes from the club's approved recipe set.

✓Return what the member logged and what the food database holds.

✓Offer the club's configured referral into a qualified.

Any write happens inside the boundaries agreed at implementation, never ahead of review.

Nutrition for pregnancy, postpartum or under-18 members.

Reading a blood result, a scan or a clinician's letter.

Adapting a plan where the state requires prior sign-off.

Changes to scope, referral thresholds or the state rules.

Example output

One suggestion, annotated

Everything the agent suggests is attached to the profile it was drawn against.

Nutrition-guidance output · single memberIllustrative example

Member

Suggested line

Basis

Drawn from

Confidence

Scope

General member

A higher-protein breakfast from the club's approved recipe set

Recipe card

Configured library

91%

General information only

As receivedTaken from the configured library and what the member logged — nothing on this side is composed by the agent.

Sources used Approved recipe set Member's logged entries Club referral list

Why this stays generalIt describes food in general, not this member's requirement — the line the dietitian weighs.

ActionReleaseEditRefer to a professional

What the score decidesBelow the configured threshold the suggestion picks up a scope review before it.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04

Every questionFrom the member

03General information

Answer from the library

Draw on the configured library and the club's approved recipe set, and on nothing else.

01Approved path

General information, named so

Routine questions about food and macronutrients arrive already answered.

02Human review

Send the rest to a person

Screened topics and low-confidence suggestions are held, so the dietitian's time goes where the risk sits.

04Build an evidence trail

The suggestion, the profile it was drawn against and the referral stay on file.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

Member managementMindbody · Glofox
ABC Fitness · Xplor

Nutrition and food dataConfigured club library
Food-composition databases

Member appsClub app · member portal
Logged food and activity

Agent

Nutrition guidance

Reads the profile
Answers generally
Holds for the dietitian

Referral and clinicalDietitian rota · referral list
Practice-management systems

Observability & evaluationOpenTelemetry · Langfuse
Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the member

Controls stack inward. What the stack does not catch is set out in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to logging only when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, library and state-rule-set changes.Track

L4TraceabilityRecord the sources, the suggestion, the screening hits and the release time.Record

L3Dietitian releaseHold suggestions for the named dietitian; it governs release, not whether the text fits its reader.Gate

L2Topic suppressionScreen each question against the suppressed-topic list; a hit stops the answer rather than improving it.Restrict

L1Confidence thresholdsRoute low-confidence suggestions to a scope review before the dietitian sees them.Require review

Model coreSuggestion produced — general information, meal ideas, scope notes and confidence

L1 – L2Test whether a suggestion may stand

L3Puts the release in a dietitian's hands

L4 – L5Keep what was suggested and what it was drawn from

L6Narrows to logging-only when signals degrade

How Nestack evaluates it

Evaluate the workflow — a correct answer in the wrong context is still a failure.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — what the member reads

Depth of coverage ▼

E1Final-output evaluationDid the suggestion stay inside the library and general framing?

E2Step-level evaluationDid the agent use the right profile, flags, library and state rule set?

E3Tool evaluationDid it read the correct member record and write the correct field?

E4Confidence calibrationDo low-confidence suggestions actually attract more dietitian edits?

E5Slice evaluationHow does performance change across specific member profiles?

E6Business outcomeHow many suggestions needed a dietitian edit or a correction after release?

Floor — the outcome the operator answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one originates.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

NU-03

Stale referral destination

A helpline on the referral list has since been disconnected.

Stage gathersLibrary, recipe set, member flags and the referral list

02 · Reasoning2 modes

NU-04

Context-blind correct answer

Standard weight-control guidance is returned to a member at risk.

NU-06

General information individualised

A population-level range is restated as this member's target.

Stage proposesGeneral information, meal ideas and confidence

03 · Tool / write2 modes

NU-02

Suppressed topic answered

A screened question is answered instead of being held.

NU-05

Product named to a tested athlete

A named supplement reaches a member subject to testing.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

NU-01

Scope line crossed

The reply reads as advice for this member's body and goals.

Stage returnsThe suggestion the dietitian releases and

05 · Change / Version1 mode

NU-07

Silent scope regression

A model or library change widens what the agent will answer.

Stage tracksModel, prompt, library and state rule set

Sev-1 · a suppressed topic is answered Sev-2 · individual advice reaches a member Sev-3 · source degrades, suggestion routes to review

See the complete Sports & Fitness failure-mode catalogue →

Affected slices

Overall accuracy can hide one member cohort

Aggregate suggestion quality can look acceptable while a small number of member cohorts absorb nearly all of the missed referrals. Nestack measures the referral-miss rate by slice, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Disordered-eating flags	6.3%	3.6×	Review
Tested athletes, supplements	4.2%	2.4×	Review
Exclusive-scope states	3.0%	1.7×	Watch
General members, no flags	1.4%	0.8×	Normal

Bar: referral-miss-rate lift vs. the general-member baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

No cycle closes without a new test

An apology does not close a cycle; a case the next release must pass does. That suite is what the next suggestion leaving the agent is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Referral-miss rate rises in a member cohort.

02Diagnose

Pull the suggestion and the profile it was drawn against, then read until one cause survives.

03Improve

The correction is versioned, with the suggestions that motivated it attached.

04Verify

A failing case holds the release until it passes.

05Learn

The case is added permanently, and the referral thresholds move with it.

Learn → DetectThe return edge. The next detection is measured against a suite that just grew.

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, guidance workflow, evaluation, integration, then production validation and handover.

Workstream Week 1Week 2Week 3Week 4Week 5Week 6

01Nutrition workflow discovery and scope definition.

02Member-system and food-data source.

03Library, referral-list and state rule-set mapping.

04Member-record ingestion and entry.

05Guidance logic and library binding.

06Confidence scoring and screened-topic routing.

07Dietitian review workflow.

08Member-app and club-system integration.

09Referral-threshold cases.

10Guardrails and scope controls.

11Suggestion-trail instrumentation.

12Deployment, documentation and Agent Care handover.

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 club, one library ProductionProduction member systems AdvancedMultiple brands / sites

Introduced at Pilot

General information from your library✓✓✓

Dietitian release✓✓✓

Suggestion-quality baseline✓✓✓

Introduced at Production

Reporting by member profile—✓✓

Release workflow in your systems—✓✓

Approved write-back—✓✓

Member-app integration—✓✓

Introduced at Advanced

Multiple state rule sets——✓

Multi-stage clinical review——✓

High member volume——✓

Multi-brand programme controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, member volume, approval controls and deployment requirements.

Separate from buildBuild 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 bringWe build with it

01Your nutrition library and approved recipe set → Library ingestion and source mappingWeek 1

02Representative member questions → Guidance baseline, scope extraction and library bindingWeek 2

03Your referral list and the professionals on it → Referral-list verification and scope-boundary definitionWeek 1

04Access to relevant APIs, feeds or exports → Member-system and food-data assessment, then integration setupWeek 2

05Suggestions that should have been a referral → Referral cases and the evaluation suiteWeek 4

06Where general information has to stop → Confidence scoring, topic screening, guardrails and scope controlsWeek 3

07Named dietitians to review suggestions → Dietitian review workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Each band covers the weeks the work really takes, so week 5 carries evaluation and launch together.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Nutrition workflow discovery, scope mapping and the state rule set W2Source integration and the guidance baseline W3Guidance workflow, screening logic and release controls W4Evaluation suite, suppressed-topic checks and failure-mode testing W5Member-app integration, a pilot cohort and targeted corrections W6A cohort of members supported under a dietitian, then handover

Reading the bandEach bar sits only on the weeks its work is named in. The fifth week genuinely carries two kinds of work.

At the end of W6Validation finishes on live members and Agent Care assumes monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Sports & Fitness AI agent

Build a nutrition agent that stays on the general side of the line.

Show us your library, your referral list and the states you operate in. A dietitian you name signs anything individual, and the workflow is built around that signature.

Discuss this agent →

Nestack Agents · Nutrition guidanceAGT-SF-04 · Agent Care available after launch