Home/Key Details

System and process facts

How a Velociti-powered private AI pilot actually works.

WonderWave scopes one sensitive workflow, runs it inside a local or customer-controlled environment, places bots under Sentinel authority, routes model work locally where required, and leaves your team with audit trails they can inspect.

Request pilot reviewRead FAQ

At a glance

The pilot is narrow by design.

One workflow is enough to prove the system: what data is in scope, which bot can act, which model route is used, who approves risky actions, and what evidence is recorded.

1.

One workflow

Examples include document intake, claim evidence review, support triage, admin summaries, internal Q&A, or client request preparation.

2.

Local boundary

Workflow data, prompts, logs, evidence, and outputs are designed to stay inside the approved environment for the private path.

3.

Sentinel authority

Sentinel reviews every bot against rules you set: purpose, target, duration, allowed tools, denied actions, and approval owners.

4.

Human gates

Write actions, exports, risky decisions, and final operational use can remain approval-gated instead of automatic.

5.

Runtime proof

Rancher-backed visibility helps show whether bot sandboxes, labels, cleanup, and runtime state match the Sentinel record.

6.

Audit trails

The pilot records approvals, denials, bot activity, model route, runtime logs, evidence files, and closeout notes.

System path

What happens before a bot touches work.

A chat request does not become uncontrolled execution. Velociti turns the request into a Sentinel-reviewed authority record first.

  • Business user or IT owner requests a bounded workflow action.
  • Sentinel checks purpose, target system, duration, filesystem access, network route, and allowed tools.
  • Approved bots inherit the Sentinel boundary and run with a lease.
  • Local model routing is recorded with the workflow evidence.
  • Runtime and GitOps evidence tie the action back to source, deployment, and cleanup state.
Pilot control pathLocal

Request

Chat request, workflow owner, target system, and purpose captured.

Sentinel gate

Rules checked before any bot, model, or runtime action begins.

Bounded bot

Bot runs under approved scope and records activity to the evidence trail.

Closeout

Approvals, denials, model route, logs, and cleanup state are retained for review.

Pilot facts

Inputs, outputs, and operating rhythm.

These are the details buyers usually need before deciding whether a private AI pilot is worth scoping.

Buyer inputsWorkflow owner, current tools, sample or redacted data, source examples, approval owner, target systems, and known risk points.
WonderWave workBoundary mapping, Sentinel rule design, local model route planning, bot workflow design, prototype build, testing, and evidence packaging.
Technical touchpointsApproved documents, folders, chat surface, local model endpoint, Rancher runtime, Forgejo/GitOps records, and output destinations.
Human reviewRisky actions, exports, final decisions, policy exceptions, and low-confidence outputs remain with named reviewers.
First milestoneA working private workflow simulation using approved examples, with Sentinel gates and reviewable logs visible to your team.
Post-pilot decisionExpand, revise, pause, or move to governance support based on workflow evidence and team review.

Timeline

A practical first pilot path.

The timeline is adjusted to the workflow, data availability, and access requirements. The important part is sequence: scope first, then build, then inspect evidence.

Discovery

Confirm workflow, stakeholders, source systems, sensitive data categories, and approval points.

Sentinel scope

Define rules for target systems, allowed tools, denied actions, lease duration, and reviewers.

Pilot build

Build the bounded bot workflow, local model route, runtime view, and evidence capture.

Review

Inspect outputs, logs, approvals, denials, runtime proof, and rollout recommendation.

Bring one workflow

The key detail is whether the work can be safely bounded.

A pilot review starts with the workflow, the local environment, the data category, the current tools, and the approvals your team needs before AI can help.

Request pilot reviewView demo