docs: explain Insider Voice, current Cat UI, and agent capabilities

[enyst]

Insider Cat uses one saved OpenHands agent behind text and Voice. Document the current Cat interface, direct realtime handoff, normal tools and enabled skills, and how the agent queries its own backend through the terminal.

• Show the current source-derived Cat artwork and poses, integrated Talk controls, and compact companion away from Projects.
• Explain realtime Voice → conversation-bound handoff → saved OpenHands tool loop → saved result → speech, including why the SDK’s stateless /ask_agent endpoint is not the task executor.
• Correct the bespoke inventory-tool prerequisite. The normal terminal, openhands-api skill, and actual backend URL/authentication context already support backend queries. Prefer the count endpoint; paginate listings; only inspect persistence when its ownership and format are known.
• Record real September 20 Voice verification: the same saved Cat invoked the API skill, discovered OpenAPI, executed a terminal count, saved the answer, and returned matching speech. An independent count agreed. The repair preserved its existing history and identity. A separate temporary profile launch verified new-Cat tools, eleven default skills, project discovery, and runtime references without starting an agent turn.
• Preserve earlier relay experiments as dated history. Physical iPad audio, progress acknowledgments, and worker mutation flows remain separate checks.

Rebased onto main 30205f5; current notes contain no private conversation IDs, credentials, or machine paths. The drawings are source-derived artwork, not claimed screenshots.

Validation: four HTML pages and 100 local links/anchors; metadata, unique IDs, privacy scan, and diff checks. Companion App tests: 129 passing. Canvas runtime-proxy tests: 88 passing plus typecheck and focused lint. Live tests used real providers with generated speech; they do not establish physical microphone/speaker behavior or the API-key Voice transport.

Related: enyst/insider#2 · OpenHands/OpenHands#17558 · OpenHands/software-agent-sdk#5180

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}