Biotechnology AI agent · Animal care
Animal Care & Use Protocol Governance AI Agent
Track each protocol, amendment and semiannual review against the interval it is measured by, and put the calendar, the findings and their correction dates in front of the committee that signs.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- IACUC approval
- Programme-wide
- After launch
- Agent Care
What this agent does
Builds the calendar, never the approval
In
- 01
A protocol is approved, and the agent runs its expiry against the 9 CFR 2.31(d)(5) floor and your own stated interval.
- 02
A review falls due, and the agent counts the programme review and the facility inspection as two six-month clocks.
Reason
- 03
An amendment is filed, and the agent holds the protocol at its current scope until the committee records a decision.
- 04
A finding is recorded, and the agent shows the plan-and-schedule date the committee wrote against today.
- 05
A correction date passes, and the agent starts the 15-business-day notice and names every federal sponsor on it.
Decide
- 06
A person joins a protocol, and the agent tracks training and occupational-health enrolment as separate states.
- 07
A fiscal year closes, and the agent derives both 1 December reports from one record and shows why the counts differ.
Out
- 08
A protocol lapses, and its animals, its findings and the assurance it sat under are retained against the record.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles the record and runs the clocks; a majority of the committee approves, and the institutional official signs the assurance and the annual report.
Example workflow
One protocol, approval to expiry
- 1Protocol receivedProtocol system, amendment form, veterinary record or facility register
- 2Record assembledSpecies, procedures, personnel, training and facility, each with its named source
- 3Calendar draftedExpiry dates, review windows, open findings and confidence
- 4Controls appliedInterval checks, amendment and training currency, finding dates and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is decided or closed at any of them — the agent is assembling, and the committee's lane opens at the confidence gate.
Goes to the IACUC chair to read.
Adds an attending veterinarian read first.
Committee review
The record is held with its sources, its open findings and the confidence.
Approve · Edit · Send to the attending vet
Approved — released to the committee
- 6Protocol registers updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedRecord completeness, finding-date outcomes, review turnaround and post-report corrections
Edits
Every committee edit is counted, and so is every correction made after the report was signed.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Writing the alternatives narrative under 9 CFR 2.31(d)(1)(ii).
- Running the alternatives search the investigator must consider.
- Determining that a procedure causes no more than slight pain.
- Assigning the USDA pain and distress column for a protocol.
Automation boundary
Agent acts unaided
- ✓Run the protocol, amendment, review and training clocks.
- ✓Assemble the evidence behind both 1 December annual reports.
- ✓Hold each finding open against the date the committee wrote.
- ✓Flag a narrative that is missing, undated or copied, and hold the record.
Any write happens inside the boundaries agreed at implementation, never ahead of a committee vote.
- Deciding a protocol, an amendment or a suspension.
- Closing a finding, or recording it as corrected.
- Signing the assurance, the annual report or a noncompliance report.
- Changes to interval, escalation or reporting rules.
Example output
One protocol record, annotated
The record is what the committee reads before it votes, and what an inspector reads afterwards.
Protocol-record output · single protocol
Illustrative example- Protocol
- Three-year rodent study
- Flag raised
- The alternatives narrative is undated and repeats an unrelated protocol word for word
- Review due
- Complete review
- Source of record
- Prior protocol file
- Confidence
- 91%
- Attribution
- Investigator and protocol ID
As received
Taken from the protocol file and the committee minutes — nothing on this side is written by the agent.
Evidence used
- Narrative date recorded
- Databases the PI listed
- Prior protocol text
Why this was flagged
We check that the narrative exists and is the investigator's own — we do not write it, and we never will.
What the score decides
Below the configured threshold the record picks up an attending vet read before the committee.
Action
ApproveEditSend to the attending vetValue
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 protocol
From the protocol file
Assemble from the record
Draw on the protocol file, the minutes, the training roster and the facility register — the record, never the deliberation.
The committee decides, not us
The chase, the clocks and the assembly come back; the committee's deliberation time is not the saving.
Aim the review at what is late
Protocols near expiry and findings past their correction date surface first; the agent never records one as corrected.
Build an evidence trail
The protocol, the review it is due and the official who signed stay on the assurance.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Protocol systems
- Cayuse
- Topaz Elements
- Key Solutions
- eSirius
Vivarium & census
- Tick@Lab
- Granite
- Animal census records
Training & occ health
- CITI Program records
- Occupational health file
Agent
Protocol & programme governance
- Reads the record
- Runs the clocks
- Holds for the committee
Facilities & findings
- Facility inspection log
- Semiannual report file
Observability & evaluation
- OpenTelemetry
- 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 committee
Six shells, outermost read first. What none of the six holds is named in the map beneath this.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a record may stand
- L3
- Leaves the vote with the committee
- L4 – L5
- Keep the protocol and the review behind it
- L6
- Reverts to calendar assembly when signals degrade
How Nestack evaluates it
Evaluate the governance workflow — not only the finished calendar.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the record the committee reads
- E1Final-output evaluationDid every date in the record trace back to the document it came from?
- E2Step-level evaluationDid the agent use the right protocol, interval and the rules in force that day?
- E3Tool evaluationDid it read and write the correct protocol and the correct field?
- E4Confidence calibrationDo low-confidence records actually attract more committee edits?
- E5Slice evaluationHow does performance change across specific protocol types?
- E6Business outcomeHow many records needed a committee edit or a correction after the report was signed?
Floor — the outcome the institution answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each set at the stage where it first arises.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- EY-03
Rescinded guidance
Reporting read from a notice rescinded in January 2026.
Stage gathers
Protocols, reviews, training records and rules
02 · Reasoning
2 modes
- EY-04
Interval read as a quota
Two reviews in the year, the six-month gap breached.
- EY-06
Annual cycle enforced
A three-year complete review is never scheduled.
Stage proposes
Expiry dates, open review windows and confidence
03 · Tool / write
2 modes
- EY-02
Amendment read as decided
Work starts while the committee has not voted.
- EY-05
Escalation list short
The notice goes out and one sponsor is missed.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- EY-01
Finding without a date
No correction date, so nothing can age against it.
Stage returns
The record the committee reads before it votes
05 · Change / Version
1 mode
- EY-07
Silent interval drift
A config change widens the window the agent accepts.
Stage tracks
Model, prompt, interval rules and reporting config
- Sev-1 · animals used on a lapsed protocol
- Sev-2 · a wrong date reaches the report
- Sev-3 · sources degrade, record goes to review
Affected slices
A programme-wide figure can hide one species group
A programme-level calendar-currency figure can look sound while a few species groups and protocol types carry most of the rework behind it. Nestack reports the record-rework rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Non-human primate protocols | 6.7% | 3.6× | Review | |
| Amended field-study protocols | 4.7% | 2.5× | Review | |
| USDA-covered farm species | 3.0% | 1.6× | Watch | |
| Single-site rodent protocols | 1.9% | 1.0× | Normal |
Bar: record-rework-rate lift vs. the single-site rodent 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 case
The loop shuts when the lapsed protocol is a regression case. That suite is what the next review calendar built is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Record-rework rate rises in a protocol slice.
- 02
Diagnose
The protocol that expired between the study and the paper is read back until the cause narrows to one.
- 03
Improve
Any change ships with a number, and the protocols attach to it.
- 04
Verify
One protocol case still failing is enough to hold the release.
- 05
Learn
One case joins the suite, one line joins the programme record.
Learn → Detect
The return edge. The next calendar is built 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, governance workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Programme workflow and escalation-boundary definition.Week 1
- 02Protocol and vivarium source assessment.Week 1, Week 2
- 03Interval, amendment and training-currency mapping.Week 1, Week 2
- 04Protocol-record ingestion and mapping.Week 2
- 05Calendar logic and finding-date binding.Week 2, Week 3
- 06Confidence scoring and record routing.Week 3
- 07Committee review workflow.Week 3, Week 4
- 08Protocol and vivarium-system integration.Week 5
- 09Calendar and currency cases.Week 4, Week 5
- 10Guardrails and reporting controls.Week 4
- 11Protocol-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 programme, one committee | ProductionProduction protocol systems | AdvancedMultiple assurances / sites |
|---|---|---|---|
| Introduced at Pilot | |||
| Clocks run to your intervals | ✓ | ✓ | ✓ |
| Committee review | ✓ | ✓ | ✓ |
| Calendar-currency baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by species group | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Protocol-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-regulator interval rules | — | — | ✓ |
| Multi-stage committee reviews | — | — | ✓ |
| High protocol volume | — | — | ✓ |
| Multi-assurance reporting controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, protocol volume, review 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 protocol list and species groups
→Protocol-record ingestion and mapping
Week 102Representative protocols and semiannual reports
→Calendar baseline, interval extraction and record binding
Week 203Your interval, escalation and reporting rule sources
→Interval, amendment and training-currency mapping
Week 104Access to relevant APIs, feeds or exports
→Protocol, vivarium and training assessment, then integration setup
Week 205Programmes you would not want inspected
→Expiry cases and the evaluation suite
Week 406What no annual report may omit
→Confidence scoring, record routing, guardrails and reporting controls
Week 307Named committee members to read records
→Committee review 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
What fixes each width is elapsed time and not layout, so the fifth band honestly carries a pair.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Programme workflow discovery, interval mapping and the escalation boundary
W2Protocol, vivarium and training integration and the calendar baseline
W3Governance workflow, confidence logic and committee-review controls
W4Evaluation suite, interval checks and failure-mode testing
W5Reporting-system integration, pilot protocols and targeted corrections
W6One semiannual cycle run under the attending veterinarian, then Agent Care handover
- Reading the band
- A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
- At the end of W6
- Once the review calendar validates, Agent Care assumes the agent.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Biotechnology AI agent
Build a governance agent that will not let a protocol lapse quietly.
Show us your protocol list, your semiannual calendar and who signs. Your IACUC coordinator already knows which protocol is closest to expiry and cannot prove it to anybody else. We sell calendar, completeness and record — not approval throughput, not faster approvals, and not less committee friction, because some of that friction is the committee doing its job. Two annual reports fall on 1 December, harmonised onto that date by NOT-OD-20-109 of 21 May 2020, and they describe two populations that will never tie: the Animal Welfare Act definition at 7 U.S.C. 2132(g) excludes purpose-bred rats and mice and birds, while PHS Policy covers all live vertebrates. We hold both from one record and show why the two counts differ. If a date we tracked is wrong after the annual report is signed, the unwind is a preliminary report, a corrective plan and a final report the institutional official signs, plus a re-examination of the tracker that failed. One refusal, stated in front: we do not write the alternatives-search narrative required by 9 CFR 2.31(d)(1)(ii). The regulation requires that the investigator considered alternatives, and the narrative is the evidence of that consideration — so automating it produces a compliant artefact of a deliberation that never happened. Our research-literature copilot, on the shelf next door, runs a recorded search, screens records against stated criteria and extracts outcomes into an evidence table; it could draft this narrative tomorrow, and we will not point it here. We will tell you the narrative exists, when it was written, which databases and date ranges the investigator recorded, and whether it was copied from another protocol. Prompt reporting to OLAW runs on NOT-OD-25-148 of 20 January 2026, which replaced the notice rescinded that day and sets no day count, so we age it and do not invent a deadline. We found no regulator-published deficiency or citation rate, so nothing here rests on one.
- Typical delivery
- 4–6 weeks
- Deployment
- Your protocol and vivarium systems
- IACUC approval
- Programme-wide
- After launch
- Agent Care available
Nestack Agents · Animal care governance
AGT-BT-20 · Agent Care available after launch