FILE // OTACON-AI-ECOSYSTEM TIER // OTACON CORE
Otacon Corefree · full source

Project 01 // Agent Platform

Otacon AI Ecosystem

The foundation of a Keep: persistent AI agents with their own identity, memory, and voice, routed through a provider-neutral core you run on your own hardware. See what a Keep is for where this is headed.

Install Otacon Install AI9 Manga Colorizer View on GitHub

Right Now

Otacon Core today

Exactly what installing Otacon gives you the moment it finishes. No roadmap items included.

✓ Available the moment install finishes

  • ✓ Install Otacon locally on your own machine, one command, and it starts itself from then on
  • ✓ Default Model included: Ollama + a chat model sized to your GPU VRAM
  • ✓ Genome Voice Trainer included for custom Piper agent voices
  • ✓ Open your own local Otacon web interface and chat with an agent
  • ✓ Create and manage persistent agents
  • ✓ Maintain separate conversation history and long-term memory per agent
  • ✓ Configure agent identities and voice profiles
  • ✓ Provider-neutral capability routing: swap LLM/TTS/STT backends without touching agent identity
Normal Otacon install = Core + Default Model + Voice Trainer. Speech-to-text and image/video still need their own providers when you want those (see status).

Architecture

Built on a resource arbiter, not a script

User→ Agent (identity · memory · voice)→ Capability Router→ LLM / TTS / STT Providers

A provider-neutral resource arbiter sits beneath capability routing: workloads request logical CPU/RAM/GPU/VRAM and receive leases that respect placement, priority, and health, rather than agents hardcoding device indexes. Idle model residency can be reclaimed gracefully while active work is protected. Per-voice synthesis settings are configured individually rather than one global profile applied to every voice. This is the same architecture that, at full scale in the private Keep, extends to media, home, and multi-node GPU orchestration. See The Keep.

Install // One Click

Install Otacon

One command. Includes Default Model (Ollama chat) and Genome Voice Trainer. Pick Windows or Linux.

Otacon

One click

Core + chat model by VRAM + Voice Trainer. Under 6GB → qwen2.5:1.5b · ≥6 → 3b · ≥8 → 7b · ≥16 → 14b.

Windows

Download OtaconsKeep Setup
These are normal — not failures

If something below happens, do not close everything and start over. Setup is still working.

  • Windows SmartScreen may say it protected your PC. Click More info, then Run anyway.
  • Administrator / Yes popup is required. Click Yes. A new Setup window may open — that is expected. The old window can stay open or say setup continues in the new one.
  • Long waits (minutes with a black window and progress text) are normal while Windows enables Linux and Otacon installs. Do not close the window.
  • Restart required does not mean install failed. Save your work, restart, then Setup should show WELCOME BACK and continue. Or double-click the same OtaconsKeep-Setup.bat again.
  • Ubuntu username / password screens are part of Windows Linux setup — not a crash, and not Proxmox.
  • Do not open localhost:5757 until the window says OTACON IS READY. Opening early looks broken even when install is fine.
  • If you see SETUP NEEDS HELP, that is recovery — not a crash. Press R to retry, O for logs, or X to exit. The window should stay open.

Saves as OtaconsKeep-Setup.bat. Double-click, follow the guided stages, restart once if asked. Setup resumes automatically. Do not open localhost until it says OTACON IS READY.

Having trouble downloading? Download the GitHub ZIP instead.

Linux · Ubuntu / Debian / WSL

curl -fsSL https://raw.githubusercontent.com/Otaconskeep/otacons-ai-ecosystem/main/install_otacon.sh | bash

First run takes a while and asks for sudo once. Registers a systemd service that survives reboots.

✓ When it's done

Otacon is ready. Open it any time at:

Open Otacon →

Chat talks to Ollama at 127.0.0.1:11434. Voice Trainer status UI (when GPU path succeeded): http://127.0.0.1:8765.

Advanced: LAN access, repair, uninstall, and other options

Use Otacon from another device on your network

By default Otacon only answers on this machine. To reach it from your phone or another computer, find this PC's LAN IP (ip addr inside Ubuntu, or your router's device list) and visit http://<that-ip>:5757 instead.

Repair

Something not starting right? Re-run OtaconsKeep-Setup.bat (or install_otacon.sh). It re-checks and continues from where you left off without wiping data. Windows status: OtaconsKeep-Setup.bat --status. Linux health: otacon doctor.

Uninstall auto-start only

Run uninstall_otacon.bat (Windows). It removes the logon task and the systemd service only. Your Ubuntu environment, install folder, and data are untouched.

Optional environment variables (set before running install_otacon.sh)

VariableDefaultWhat it changes
OTACON_INSTALL_DIR~/otacon-ai-ecosystemWhere Otacon gets installed
OTACON_BUILD_NATIVE1Set to 0 to skip building the native .deb app
OTACON_INSTALL_DEB0Set to 1 to also install the built .deb automatically
OTACON_LAUNCH_WIZARD1Set to 0 to skip auto-launching the service at the end
OTACON_INSTALL_STT0Set to 1 to install optional Faster-Whisper speech-to-text
OTACON_RUN_TESTS1Set to 0 to skip the self-test suite (faster, less safe)
OTACON_INSTALL_VOICE_TRAINER1Set to 0 to skip Genome Voice Trainer
OTACON_INSTALL_DEFAULT_MODEL1Set to 0 to skip Ollama + default chat model
OTACON_LLM_MODEL(auto)Override the VRAM-picked Ollama model tag (e.g. qwen2.5:7b)
Requires sudo to install a handful of system packages (build tools, Python, WebKit libs) and to register the systemd service. The script explicitly refuses to run as the root user itself.
Default Model (included) installs Ollama with a VRAM-tier chat model: qwen2.5:1.5b (CPU/<6GB), 3b (≥6GB), 7b (≥8GB), 14b (≥16GB). Skip with OTACON_INSTALL_DEFAULT_MODEL=0 / OTACON_INSTALL_VOICE_TRAINER=0.

Full install documentation on GitHub →

Voice Training

Genome Voice Trainer — included with Otacon

Custom Piper voices for your agents. The normal Otacon install runs this for you. Standalone only if you skipped it or need to repair.

Standalone Voice Trainer (advanced)

Windows

Download Windows Installer

Linux · Ubuntu / Debian / WSL

curl -fsSL https://raw.githubusercontent.com/Otaconskeep/otacon-voice-trainer/main/install_voice_trainer.sh | bash

Write-up: docs/WRITEUP.md

Status

What's real, what needs a provider, what's Keep-only

Straight from the repository's own status table. "Requires provider" means the routing and architecture exist and are testable today, you connect the backend. "Keep Reference" means it exists only in Antonio's private environment, not in this public repo at any maturity level yet.

Available now Requires provider Keep reference Planned

Persistent conversations

Available now

Long-term memory

Available now

Agent identity & voice profile config

Available now

Native installer & hardware planning

Available now

Backup / restore / safe uninstall

Available now

Agent chat (Default Model / Ollama)

Included with Otacon

Voice synthesis (Piper)

Requires provider

Genome Voice Trainer (GPU Piper)

Included with Otacon

Speech-to-text

Requires provider

Image / video generation

Requires provider

Single remote compute node

Requires provider

Agent rooms

Keep reference

Media automation, Home Assistant, search

Keep reference

Multi-node GPU orchestration

Keep reference

Discord / messaging integrations

Planned

Three Tiers

Otacon Core, the Keep Blueprint, and Otaconskeep Services

TierWhat it isStatusPrice
Otacon CoreFree · open source · self-hostedInstall todayFree
Keep BlueprintThe broader ecosystem architecture demonstrated by the real KeepDirection, not yet a public installerNot offered publicly yet
Otaconskeep ServicesArchitecture · deployment · supportAvailable nowContact for scope/pricing

There's no license key or artificial limit in Otacon Core. It's the real, complete public platform as it stands today. Otacon Core installs the foundation. The Keep Blueprint shows what that foundation can become.

💬 Ask in Discord