arizuko › products
products
Each product is a folder shape — a persona, a skill set, the
channel wiring, and an operator setup guide. Some ship as templates
(arizuko create --product trip, reality,
slack-team); the rest are a short recipe of CLI commands.
Either way you end up with files you edit.
Under the hood a product is no new machinery: it’s the same six-primitive pipeline every arizuko agent runs, with the folder contents and routing fixed for one job. That’s why these are yours to read, diff, and change — a product you own, not a blob you rent.
Products span a spectrum: a single shared folder with one job at one end, a full org-chart hierarchy at the other. Pick the shape that fits, then own the files.
Supermarket price agent
One folder, one catch-all route, no per-user state. Answers “where’s X cheapest?” for any number of users without changing structure. The minimal end of the spectrum: same code serves 10 or 10,000 users — the limit is container slots, not registered users.
Trip planner (May)
State the brief once — May researches, synthesises via oracle, and delivers a day-by-day itinerary. Preferences persist across trips. Mid-point on the spectrum: single-user depth without an org hierarchy.
Slack team agent
One agent per Slack channel: channel-scoped persona and thread context, per-teammate memory and grants, personal cloud storage, web interface, and usage dashboard. OAuth sign-in links a teammate’s identities across chat and web.
Discord team agent
One agent per Discord guild: channel-scoped persona, per-member memory and grants, personal cloud storage, web interface, and usage dashboard. Use cases include support agents (KB-backed Q&A with citations) and collaborative research (news ingestion, observation tracking, synthesis).
AWS DevOps agent (Argus)
An SRE in your team’s chat that reads your runbooks and drives AWS read-before-write. Each engineer acts with their own keys, so CloudTrail names the human, not a shared robot. Default-deny egress means the agent can’t reach anything but the hosts you allowlist. Per-operator credentials are the differentiator most AI-SRE tools skip.
Reality agent (Rhias)
Long-running personal agent. Holds named threads for ongoing situations across weeks; structured reflection on demand; monthly scheduled sweep over diary + threads to surface what changed.
Company brain
A network of per-team agents with their own inboxes, knowledge bases, and scheduled briefs. Folder tree mirrors the org chart. Intake from Slack, email, and webhooks. The complex end of the spectrum: all six primitives at full depth.
Taxonomy chat (Marble)
A read-only chat over a fixed dataset that ships inside the folder. The agent answers by running jq over bundled JSON — no web-search egress, no per-user state. Runs on a cheap model with trimmed skills. Swap the dataset and the same bot answers over your data.
See getting started for the platform install before picking a product.