Sobani is an AI roommate for your Windows desktop. One Resident connects a body, the selected Character Pack, memories you can inspect, and promises across later app starts. Life with the standard Resident is Free; saving and applying your own Character Pack is Pro.
Before you start: on BOOTH, Free is 0 JPY and Character Studio Pro is a one-time 1,500 JPY purchase for Windows 11. For conversation AI, prepare one route: Ollama plus a compatible model, or an API key and available model for OpenAI, Anthropic, or Gemini. Ollama is not required for a cloud route.
From the day you install Sobani, she becomes a resident of your desktop. She sits on window edges, takes an occasional walk, and spends quiet time simply being there.
She does not look at your screen without permission. Screen awareness and PC control only work when you explicitly enable them.
In chat, Sobani passes basic identity, personality, speech, relationship, background and worldview, and conversation examples to your chosen Conversation AI as the selected Character Pack. While you are not chatting, that same AI uses the pack and current facts to decide whether to speak, stay silent, or use a body-mechanics intent. You can set the contact cadence.
By default, Sobani connects to a local Ollama model you install separately. OpenAI, Anthropic, and Gemini APIs are optional choices.
Your completed personal conversations are first retained as events on this PC. From them, your selected Conversation AI organizes low-risk facts and preferences that may matter later, and a separate check must pass before they become memories.
Not every utterance becomes a permanent memory, and you do not need to say “remember this.” You can inspect, approve, exclude, edit, or delete memories and review candidates. Later conversations use those same records.
Ask during a conversation and Sobani writes it on the promise board. If the app is running at that time, she reminds you in her own words. One-time, daily, and weekly promises are supported.
Promises are stored only on this device. Sobani cannot notify you while the app is closed.
One Sobani capability / Included in Free and Pro
This capability is called “Stage Beta” inside the app. While Sobani is running, the currently applied character can join allowed text channels or a selected voice channel through a Discord bot you create and invite. The same character can also appear in a transparent OBS Browser Source served only from this PC.
The bot receives mentions and replies in allowed text channels. In a selected VC, voice conversation starts only after both mandatory notices are delivered. It is online only while Sobani is running and residency is enabled.
The applied Character Pack is passed to the Conversation AI for each reply. Stage traffic remains in its local utterance log and does not enter your personal Brain memories. Daily reply and token limits are on by default.
Show the currently applied Resident in a transparent OBS Browser Source. Its tokenized URL is an access key; never expose it on stream or to another person.
Free uses the standard Resident; Pro can also use the currently applied custom Character Pack. Version 1.2.13 supports Discord text participation, VC voice conversation, and OBS display. Before listening, VC announces “an AI is listening” in both text and voice; if either notice cannot be delivered, the join is aborted. It does not support 24/7 residency or Discord-originated control of your PC, files, or settings.
Setup requires your own Discord bot, an invitation to your server, Message Content Intent, and a bot token. Windows encrypts the token at rest, and Sobani uses it directly to authenticate with Discord; there is no Sobani-operated relay bot. VC also requires separately installed local whisper speech recognition and a running VOICEVOX instance. Audio is transcribed only on this PC. Whisper and VOICEVOX are not bundled. If you select a cloud conversation-AI provider, Discord text or the transcript and the context needed to generate a reply are sent to that provider and may incur API fees.
Chat history, memories, promises, and Character Pack settings are stored on your PC. With Ollama, conversation and memory cultivation are processed locally too.
When you select OpenAI, Anthropic, or Gemini, by default that same provider and model handle replies and organize completed personal conversations (a separate Memory AI can be chosen in Settings). There is no separate Memory ON/OFF. Organization uses up to three conditional stages; bounded retries and calls whose results are not received may still incur charges. Sobani never switches to another AI, model, or Ollama. API keys use Windows encryption.
You can delete individual memories, disable permissions such as screen awareness at any time, and remove the app's local data folder.
| OS | Windows 11 (64-bit) |
|---|---|
| Memory | 16 GB or more recommended for running a local LLM |
| Required | One conversation-AI route: Ollama (free) with a compatible model, or an API key and available model for OpenAI, Anthropic, or Gemini. Ollama is not required for a cloud route. The included README explains the setup. |
| For Stage Beta | Your own Discord bot, a server invitation, Message Content Intent, and a bot token. VC also needs separately installed local whisper speech recognition and a running VOICEVOX instance. |
| Optional | VOICEVOX (free, installed separately and not bundled) — spoken lines use VOICEVOX: Shikoku Metan. Stage VC uses it for the notice and replies. |
| Disk | Space for the app. If you choose Ollama, allow several additional gigabytes or more for the model. |
Available now on BOOTH
Free delivers the finished Sobani Resident experience. Character Studio lets you inspect your VRM and .vrma motions, then edit, validate, and preview basic identity, personality, speech, relationship, background and worldview, and conversation examples as one Character Pack.
Free
Use Resident features such as conversation, memories, and promises with the standard Resident persona, Mira. In Character Studio, you can edit, validate, and preview a Character Pack and resume an unapplied draft.
Free cannot save a reusable custom Character Pack or apply it as the active Resident.
0 JPY (Free variation)
Character Studio Pro
Save the character you create in Character Studio as one reusable Character Pack and apply that pack to the Resident experience.
1,500 JPY one time (Pro variation)
Existing BOOTH or itch.io purchasers can use Pro without paying again or losing features.
Safety, privacy, data inspection, and data deletion remain available in Free.
BOOTH item 8616433 is live with Version 1.2.13 Free at 0 JPY and Character Studio Pro at 1,500 JPY.
Internet access is needed when the initial setup downloads extra data. With the Ollama route, install Ollama and a compatible model; conversations can then be processed on your PC. OpenAI, Anthropic, and Gemini routes need internet access during conversation. VOICEVOX is an optional separate install.
Character Studio lets you inspect your VRM and .vrma motions, then edit, validate, and preview basic identity, personality, speech, relationship, background and worldview, and conversation examples as one Character Pack. Saving and applying a reusable custom pack requires Pro.
It is one Sobani capability. While Sobani is running, your own Discord bot can join allowed text channels or a selected VC, and the same Resident can appear in a local OBS Browser Source. Version 1.2.13 VC uses separately installed whisper for on-PC transcription and starts listening only after both the text and spoken notices are delivered. Spoken replies require VOICEVOX to be running.
With Ollama, conversation and memory organization stay on this PC. If you select OpenAI, Anthropic, or Gemini, by default the same provider and model handle replies and post-turn organization (a separate Memory AI can be chosen in Settings); there is no separate Memory ON/OFF. Stage 1 sends the completed owner utterance and stage 2 sends candidates with their evidence. When stage 3 is needed, it compares candidates with one another and with up to 24 current low-risk owner memories of the same type. Pending-review-only candidates skip stage 3, and pending-review, sensitive, and protected memories are excluded from it. Retries and calls whose results are not received may still incur charges; Sobani never switches to another AI, model, or Ollama. For every completed owner conversation that is saved, regardless of the first reply generation's classification, one dedicated source-attestation check sends only the full current user message to the same provider and model. It does not send assertion quotes proposed by the first generation, conversation history, existing memories, or the proposed reply. The checker independently returns exact spans for the complete assertion, its subject, and its value from that message. An ordinary cloud conversation makes two calls; when organized memory is used in the reply, memory selection, reply generation, semantic consistency checking, and the pre-save source-attestation check make up to four calls normally. If the reply is rejected, Sobani does not regenerate, display, or save it. Those reply calls are separate from the post-turn stages.
Install the free VOICEVOX app separately and Sobani can read her lines with VOICEVOX: Shikoku Metan. VOICEVOX is not bundled. Text-only conversation works without it.
Free is 0 JPY and Character Studio Pro is a one-time 1,500 JPY purchase. There is no subscription. If you choose cloud AI, you pay that provider separately.
Invite her to stay.
Free on BOOTHFree is 0 JPY and Character Studio Pro is a one-time 1,500 JPY on BOOTH for Windows 11. Prepare one conversation-AI route: Ollama plus a compatible model, or a supported cloud API.