Kotoba

Where to start

What Kotoba isWho she is and where she runs, in one page
InstallingOne package, two commands
First runA key, a model, her first words
The approval gateHow she asks before she acts
The two voice modesLocal voice, or the agent tunnel
The soul fileChange who she is
RoadmapWhat grows next, and what was cut on purpose

Or a section

↑↓ move openesc close124 pages
All pages

Contributing and reference9 of 12

Roadmap, and what is deliberately not planned

About 3 minutes to read

On this page

Version 0.10 is the first release: she runs on your machine, talks, and does things. This page is the honest list of what she cannot do yet, and the shorter list of what she is never going to do.

No dates are promised. The order below is roughly the order of work, and that is the strongest claim anyone here can make about it.

What the first release already gives you

BrainsOpenAI and xAI, chosen at setup and switchable later
VoiceElevenLabs, in and out, with two engines to pick between
Avatara Live2D model of your own; the browser's first-run screen fetches the free sample for you
Surfacesa browser app, a full terminal, and a Discord bot that reads and talks
Toolsshell, code, files, web, memory, vision, reports, skills, cron, subagents, MCP servers
Windowssetup, doctor, serve, discord and --once run natively

Next

Pick a voice from a list. Today you paste an ElevenLabs voice ID into Settings, which means leaving the app to go and find one. It should be a dropdown of the voices your key can actually use.

More brains, by name. Any OpenAI-compatible endpoint already works: Settings takes a base URL and a model name typed by hand. What is missing is the other half — providers listed by name, with their models and what each one costs, the way OpenAI and xAI are. Adding one means editing Python in three places.

A voice that costs nothing. ElevenLabs is required today, and that is the single biggest thing standing between Kotoba and someone who just wants to try her. A local speech engine, running on your own machine, removes it.

The full terminal on Windows. The commands run there now; the interactive session does not — it needs a POSIX terminal. Making it native is a piece of work, not a patch.

Saying no to one command. You can already switch a whole toolset off in Settings — turn off the terminal and shell is never offered and never asked about. What you cannot do is the narrow version: "never run rm, don't ask", while keeping the rest of the toolset.

A desktop app. One window, no browser tab, no terminal.

Not planned

These are deliberate absences, written down so nobody waits for them.

A hosted version you sign into. Kotoba is self-hosted. Your keys and your conversations stay on your machine. There is no account and no server run by anybody else.

Shipping a Live2D model in the repository. Models belong to their authors. The first-run screen fetches one onto your machine instead. This is not a gap waiting to be closed — see the licence page.

Telemetry. Not now, not later.

Two things a reader should know about this list

It is a list of absences, not a plan of record. Nothing here has a milestone, an owner or a date attached to it in the repository, and there is no issue tracker state feeding it. If you need one of these things by a particular week, the honest answer is that nobody can give you that.

"Next" is not "soon". The order reflects intent at the time the file was written. Check the changelog for what has actually moved.

Asking for something

Open an issue with the feature request template. What helps most is the shape of the thing you were trying to do when you hit the wall — not the feature you designed to fix it.