Proof path
Kitsoki's product claim is falsifiable: the workflow should stay in charge while the model works inside declared boundaries. This page is the short path through the demos and docs that answer that claim.
1. Can the runtime control an agent?
Watch the host reject a model submission, inject a nudge, and accept the fixed result inside the same bounded call. Then watch an agent question block on a real operator answer instead of silently defaulting.
Agent action transcripts
Every agent call's full tool-use stream — captured once, replayed byte-identically.
Operator questions, forwarded
Agent questions become blocking operator modals — never silent defaults.
2. Can I audit and replay the run?
Trace pages show decisions, confidence, evidence, host calls, world mutation, latency, and replay affordances from the recorded trace. The useful question is not "was there a transcript?" but "can I point at the decision edge that moved the workflow?"
Decision-first trace detail
Verdicts, confidence, evidence, and replay — on a frozen trace.
Trace introspection, live
View modes, waterfall, decision-first detail, annotation, replay.
3. Can this drive real repo work?
These demos show Kitsoki as a graph of rooms and gates over a real engineering workflow: triage, reproduce, propose, implement, test, review, validate, and open the PR.
Bug triage to autonomous fix
Triage a ticket, hand it to the pipeline, watch it land.
Fix a real bug in slidey, end to end
kitsoki's bug-fix pipeline reproduces, fixes, tests and validates a real slidey bug — a grid-cards narration-timing drift.
Open the PR for the slidey fix
kitsoki's PR-refinement pipeline opens, monitors and merges a pull request for the real slidey grid-cards narration-drift fix.
4. Can I try it locally?
The install path starts from your repository, not from a Kitsoki checkout. Use the binary, run onboarding from the project root, inspect what it discovered, then commit the small setup it writes.
The stranger install — download, run, onboarded
A released binary, dropped into a foreign repo, profiles it and installs the toolkit without a checkout.
kitsoki doctor — provider, toolkit, and MCP readiness at a glance
One command reports whether your provider, agent toolkit, and MCP registration are actually ready.
Read next
- Evaluate Kitsoki for the comparison against coding agents, graph orchestrators, durable workflow engines, and scripts.
- Getting started for the first local run.
- Demos and features for the full catalog.