arizuko › concepts

concepts

Each concepts page explains one idea of arizuko. The list below runs from the primitives to the address format. Complete option tables are in reference, and setup commands are in how-to.

Primitives first, then one page per idea

1. Primitives: the whole system on one page. Six parts in a fixed order, and how every feature combines them. Read it first.
2. The store: where every row lives. One server, one account per instance, one key per daemon.
3. Scaling: the same code runs solo/inbox and corp/eng/sre/oncall. A new tenant is rows in the grants, routes, secrets and tasks tables plus a folder under groups/.
4. The ant: the agent you talk to, a folder with a persona, skills, memory and secrets.
5. A turn, end to end: one figure of the whole path from the adapter through the router and the guest back to the chat.
6. Routing: how an inbound message finds the correct agent, through one table that maps JIDs to folders.
7. Engagement: after a message arrives, whether the bot keeps listening after one mention.
8. Proactive: the other direction. The agent speaks when nobody addressed it, behind a row of checks and a 24-hour cooldown. It is off until you turn it on.
9. Topics: one strand of work as its own conversation inside one folder.
10. Live progress: one temporary checklist line that arizuko posts where the answer will go and deletes when the task ends.
11. Retries: what happens when the guest that runs your turn stops before it answers.
12. Onboarding: how a new group and its members get in: invites, the admission queue and SetupGroup.
13. Autoviv: how a new chat gets its own folder and agent on first contact, with no manual step.
14. Personas: the voice and standing instructions of an agent.
15. Auth: how arizuko knows who calls: OAuth, local login, account linking and one canonical sub.
16. Grants: what that principal can do, read from the acl table on every call.
17. Scopes: the folder tree that grants use, which folder sees which, and where isolation starts.
18. Secrets: model credentials in the environment of the guest, and capability credentials used on the host.
19. Connectors: the API of another service as an agent tool, with the credential kept on the host.
20. Audit trail: the row that every change writes in the same transaction as the change, and which daemon keeps it.
21. Skills: markdown instructions that the agent loads when it needs them.
22. Tasks: scheduled and recurring runs: cron, interval, one-time and autonomous work.
23. Workflows: the discipline that makes jobs with many steps converge: free conversation, a strict opening and one plan file.
24. Tokens: route tokens that let a public chat box or a webhook reach an agent without a login.
25. Web-native agents: the web content that a folder owns: a public page, an embedded chat and form intake.
26. WebDAV: the file workspace that you can mount with rclone or reach with curl.
27. Voice: voice in and out. Whisper transcribes, and the agent decides when to speak back.
28. Slack pane: the Slack assistant sidebar with a title, suggested prompts and pane context.
29. Addressing (JID): the one address shape for every chat and sender on every adapter.