Delegate the work. Keep the evidence.
Kivi is a local-first workbench for delegated data, AI, and ML work. Give it an objective and clear boundaries. It is designed to carry the work through investigation, execution, verification, and recovery, keeping the plan, approvals, and evidence together.
Work
Objective-led
from intent to verified outcome
Product
Local-first
your project, your workbench
Scope
Data, AI & ML
one governed execution path
example objective
Investigate rising pipeline cost
Define the objective
outcome, acceptance criteria, limits
Investigate and propose
keep evidence and changes connected
Approve and execute
run within the agreed boundaries
Verify the result
check the outcome and retain evidence
status
In development
platform
macOS first
scope
Data, AI & ML
01 / The workflow
One objective, all the way through.
An illustrative workflow for the product in development: define the job, inspect the proposal, approve within limits, and check the result. Evidence stays connected throughout.
ask
“Investigate why this pipeline became expensive, propose a fix, validate it, and prepare a release.”
review path
- 1
define success and limits
keep the objective and budget explicit
- 2
investigate and propose a change
retain the diagnosis and supporting evidence
- 3
review before consequential execution
bind approval to the proposed work
- 4
verify and prepare the release
check acceptance criteria before completion
evidence to retain
The diagnosis, proposed diff, approval, run record, and validation results belong to the same objective. Kivi is designed to keep that context available when work pauses or the model changes.
Release boundary: approval required(before promotion)
Review the evidence and proposed revision before release.
Delegated work needs an objective, clear boundaries, and evidence that lasts.
02 / What it does
Six principles, one workbench.
Kivi is being built around the full lifecycle of delegated work. These principles guide the product and the capabilities being developed.
Start with an objective
Define the outcome, acceptance criteria, and boundaries before delegating. The work stays anchored to what you asked for.
Keep the work connected
The design carries plans, approvals, execution records, and evidence across conversations, model changes, and interruptions.
Set explicit boundaries
Permissions, data disclosure, and cost limits govern execution. Sensitive actions require the right approval.
Work across disciplines
Bring analytics, pipelines, applied ML, AI tools, and operations into the same project and review process.
Verify the outcome
Connect checks to acceptance criteria and retain the evidence needed to judge whether the objective was met.
Build for recovery
Preserve waiting work and reconcile uncertain operations before retrying, so an interruption does not mean starting over.
03 / Where it fits
Where it earns its keep.
The product spans data, AI, and ML workflows where a useful answer needs to become inspectable, executable work.
Data investigation
Investigate a metric change, inspect the relevant data, and keep the queries and evidence alongside the diagnosis.
Engineering changes
Propose a pipeline or model change, review the diff, validate it, and prepare the work for release within agreed limits.
Applied ML
Connect feature work, training, evaluation, and prediction to a clear objective and inspectable results.
Operational follow-through
Carry investigations into cost, failed runs, and incidents through proposed fixes and verification, with approvals along the way.
04 / Follow the work
Kivi is taking shape.
In development, with macOS first. The focus is on keeping delegated work continuous, bounded, and verifiable. Read the case study for the product choices, or get in touch to discuss how Kivi could fit your data, AI, and ML workflows.
01 / Design focus
Keep objectives, approvals, and evidence connected across sessions
02 / Design focus
Bound delegated work with explicit permissions and budgets
03 / Design focus
Verify outcomes before treating the work as complete