Otōto 2026.19

Otōto 2026.19, released as an update via `ototo update`, lets the tool be installed and used before any small model server exists, giving Claude only the model-free tools `search`, `read`, `outline`, `changes`, `history` and `replace_all` while withholding `ask`, `locate`, `callers` and `edit` until a server is available. Adding a model later takes one command — `ototo init` or `ototo init --base-url <url>` — which restores those four tools to Claude's list, and `ototo doctor` reports "no small model set up" with nothing to fix when none is configured. The release also makes `ototo ui` re-read `config.toml` each time the page opens and at "Probe now" instead of only at startup, and labels delegated versus direct calls in the dashboard and chart legend.

- Otōto without a small model is a setup of its own. You can install Otōto before you have a model server. When ototo init finds none it says so, and sets Otōto up for that: Claude is given search , read , outline , changes , history and replace all , which need no model, and is told to find code with them. ask , locate , callers and edit are left off its list, where until now they were offered and could only fail. ototo doctor shows "no small model set up", with nothing to fix. OpenCode is set up the same way. - Adding the model later is one command. Once a model server runs, ototo init it looks on this machine's usual ports or ototo init --base-url <url puts ask , locate , callers and edit on Claude's list and rewrites what Claude is told. If a server is put in config.toml by hand and that is not done, ototo doctor says so. - The dashboard shows your settings as the file says now. ototo ui read config.toml when it started, and it runs for weeks: an edit did not show until it was restarted. It now reads the file each time the page opens and at "Probe now", and probes the model servers the file names then. The settings have a panel of their own, beside the model servers. - The dashboard says what delegated and direct calls are. A line under its first figures: a delegated call is a question handed to the small model, and a direct one is answered by Otōto at once, with no model. The chart's legend says which is which, and so does each tool's label when the pointer is over it. Upgrading from 2026.18: ototo update . With a model server set up, it changes nothing in settings.json or CLAUDE.md. With none, it rewrites Otōto's block in CLAUDE.md and its hooks in settings.json for the setup described above, backing both up first.