How we ship

How a system is designed and shipped.

Constraints, then the Knowledge Image, then evidence queries, then approval-scoped tools, then eval, deploy, operate. Work is scoped privately. There is no public rate card on this domain.

Commercial path

Work starts on HIVE, or it arrives scoped.

Operators who already know they need a builder send an engineering brief here. Everyone else starts a HIVE Bootcamp. This company does not run discovery theater or free demos.

Commercial start lives at hiveai.tech/bootcamp. Scope, environment, and commercial terms are written privately. We do not publish dollars, retainers, or “engagements start at” on this domain.

Start a HIVE Bootcamp

Opened instrument case with a Knolo Knowledge Image on an engineering bench. — BAY · IMAGE OPENED · CONSTRAINTS ON THE TABLE

Constraints table: handwritten cards, sorted folders, a ruler, and a closed Knowledge Image case. — CONSTRAINTS · DATA CLASSES ON THE TABLE · SUCCESS METRIC WRITTEN

01 / CONSTRAINTS

Write the bars before the image

Name the workflow that must be true, not the model that sounds impressive.
Classify the data: public, internal, restricted, regulated. Write who may retrieve it.
Write the success metric in a sentence a counsel and an operator can both accept.
Choose the environment: on-prem, VPC, or isolated tenancy. This is not postponed.
Refuse the work if the buyer wants hours, a chatbot package, or an unevaluable demo.

02 / PACK / IMAGE

Mount a Knowledge Image

Build a portable .knolo file under class, owner, and retention.
Unclassified material does not get imaged. Smaller and clean beats large and contaminated.
The image can be hashed, shipped, revoked, and run locally.
Pack versions move through the same change control as code.

03 / EVIDENCE QUERIES

If it cannot cite, it does not answer

Queries hit the image under policy: lexical grounding, spans, plan hash, read receipt.
The same question, same image, same policy is replayable.
Embeddings, if used, sit behind that contract. They are not the trust root.
Stand the Workspace against real sources, not a slide corpus.

04 / APPROVAL-SCOPED TOOLS

Default-deny is the control

Wire the first action behind a human approval or a named capability record.
A tool call is a permissioned function with a schema and a side-effect policy.
Available actions stay deliberately narrow and require explicit authority.
Exports do not leave because a model asked.

05 / EVAL

A gate, not a feeling

Groundedness, citation coverage, tool-policy, frozen regression set.
If it cannot be evaluated, it is not in production.
Leave Bootcamp with a working path, a written gap list, and a production scope — or a stop.

06 / DEPLOY

The runtime follows the constraint

Identity: SSO, roles, least privilege on retrieval and tools.
Environment: VPC, on-prem, or isolated tenancy named in the statement of work.
Runbooks: fail closed, on-call path, change control.
Load, latency, and the boring integration work that decides whether the system survives a Monday.

07 / OPERATE

Do not abandon the system

Eval against the frozen question set plus new failures.
Change control on images, prompts, tools, and models.
An accountable operator. Hive Forensics AI can run it; we do not vanish after launch.
Expansion is a new constraint, not an unbounded backlog of “also make it do X.”

What we refuse

The engagement is accountable for a working system.

No hour rental

Hive Forensics AI deploys systems. If you need contractors on a Jira board, hire a staff shop.

No chatbot SKU

Voice agents and chat widgets are not the offer. An operator system may include a channel. The system is the product.

No unevaluable launch

If there is no eval, there is no production. We will stop before we ship a costume.

Engage

Work starts on HIVE Bootcamp.

Unscoped work starts at hiveai.tech/bootcamp. This domain is for engineering already constrained. Work is scoped privately.

Start a HIVE Bootcamp Send an engineering brief

Scoped privately · no public rate card on this domain