FAQ
Do I have to configure all this by hand?
Section titled “Do I have to configure all this by hand?”No, and mostly you should not. Yachiyo ships with the yachiyo-help skill
enabled — documentation of its own CLI, config, and setup procedures — plus shell
access. So it can administer itself: set up launch-on-login, repair your PATH,
create a schedule, curate a model list, write a skill.
Ask in plain language and check what it did. These docs exist for when something breaks, when you want the underlying detail, or when you need it scripted.
Is there a Windows or Linux build?
Section titled “Is there a Windows or Linux build?”No, and none is planned. Yachiyo is built on a Mac and there is no hardware here to test the other platforms on, so porting it is not on the table. Several bundled skills — window automation, Ghostty control, screenshots — are macOS-only regardless.
Does Yachiyo support MCP?
Section titled “Does Yachiyo support MCP?”No, deliberately. Skills are Markdown files instead: no protocol, no server process, no manifest. If you can write a README you can extend Yachiyo. That is the whole trade — you give up an ecosystem’s worth of prebuilt integrations and get something you can read, edit, and reason about.
Where is my data?
Section titled “Where is my data?”~/.yachiyo, on your machine. Conversations are in a local SQLite database.
There is no hosted backend and no account. See
Files and paths.
Is there telemetry?
Section titled “Is there telemetry?”No.
Can I use a local model?
Section titled “Can I use a local model?”Yes. The Ollama preset points at http://localhost:11434/v1 and needs no key.
Any OpenAI-compatible server works the same way. See
Providers.
Can I use several providers at once?
Section titled “Can I use several providers at once?”Yes, and it is the expected setup. Model choice is per message, not per app — draft with something cheap, escalate the hard question, keep the whole exchange in one thread.
Do I have to pick one model for everything?
Section titled “Do I have to pick one model for everything?”No. There is a chat model and a separate tool model for background work like thread titles and memory distillation. Pointing the tool model at something small and fast is the usual setup.
Do my Claude Code / Codex skills work?
Section titled “Do my Claude Code / Codex skills work?”Yes. Yachiyo scans ~/.claude/skills, ~/.codex/skills, and ~/.agents/skills
alongside its own, in both your home directory and each registered workspace.
They are discovered but not auto-enabled — switch them on in Settings →
Capabilities → Skills.
Why did my edit to a bundled skill disappear?
Section titled “Why did my edit to a bundled skill disappear?”~/.yachiyo/skills/core/ is re-extracted on every app launch. Put your own work
in ~/.yachiyo/skills/custom/ instead.
Do schedules run when the app is closed?
Section titled “Do schedules run when the app is closed?”No. Schedules, channel bots, and yachiyo send all need the app running. Set up
a LaunchAgent if you want it always on.
Can other people use my assistant through Telegram?
Section titled “Can other people use my assistant through Telegram?”Only if you let them. Users register on first contact as pending and stay
blocked until you allow them. Guests get a sandboxed workspace, filtered memory,
their own profile, and a token quota. See Channels.
Will the bot reply to everything in a group chat?
Section titled “Will the bot reply to everything in a group chat?”No. Group mode runs a cheap probe model that watches the conversation and only wakes the full model when there is something worth saying, backing off through dormant, active, and engaged phases. See group discussion mode.
Can I stop it from editing files?
Section titled “Can I stop it from editing files?”Yes. Use Explore or Chat run mode, which removes the editing tools from the model’s tool list entirely, or just do not give the thread a workspace. Every run that does touch files leaves a reviewable diff.
Are my API keys safe?
Section titled “Are my API keys safe?”They are stored in plain text in config.toml, on your machine. CLI output
redacts them, but the file does not encrypt them. Do not commit it, and be aware
that enabling sync copies it into your sync folder.
Channel bot tokens in channels.toml are not redacted in output at all.
What happens when a conversation gets too long?
Section titled “What happens when a conversation gets too long?”Older context is compacted rather than dropped — a handoff summary preserves the
thread’s substance while freeing the window. The threshold is configurable via
chat.stripCompactThresholdTokens.
Can I search old conversations?
Section titled “Can I search old conversations?”Yes, in the app or from the CLI. The full-text index is trigram-based, so it handles CJK as well as English:
yachiyo thread search "that thing about retries" --limit 10Does sync mean I can chat from any device?
Section titled “Does sync mean I can chat from any device?”Partly. Settings sync both ways, but chat archives from other devices arrive read-only — you can read a laptop conversation on your desktop, not continue it there.
Can I contribute?
Section titled “Can I contribute?”Bug fixes and documentation improvements are welcome. Feature PRs are not accepted unless the feature was discussed and approved in an issue first — open the issue before writing code. See the repository.