N.E.K.O.: An Open Source AI Companion That Talks, Sees, and Remembers You

What N.E.K.O. Actually Is

N.E.K.O. stands for Networked Emotional Knowing Organism, and the README is explicit about what it is not: not a task automation tool, not a roleplay skin, and not a cloud black box. It positions itself as an end-to-end AI companion platform that combines real-time voice and text chat, visual understanding of your screen, a five-tier memory system, and agent tool execution. The target user is someone who wants a persistent digital companion that reaches out first, shares media, and can act on their behalf, rather than a chatbot that only responds when prompted. The Steam page and the free listing suggest an audience comfortable with desktop apps and gaming, but the core is a Python project that runs locally with an Apache-2.0 license.

The Active Proactive Mechanism

The central design choice is proactivity. The README describes a companion that initiates contact based on screen understanding, social media trends, personal updates, and music memes. This is not a passive chat window. It has real-time voice via the Realtime API and text via ChatCompletion, plus visual understanding of what is on your screen. The memory system is split into five dimensions: working, recent, factual, reflective, and personality memory. The claim is that this lets the companion remember you across desktop, mobile, and gaming contexts, with full synchronization. The agent component can operate a browser and computer, calling CUA or OpenClaw A2A to complete tasks. That is a heavy stack, and the README admits the purpose is relationship, not task completion, which is a meaningful distinction for engineers evaluating it as a tool.

Running It: One-Click and Docker Paths

For Windows, macOS, and Linux, the README says to download a one-click package and run N.E.K.O.exe, N.E.K.O.app, or n.e.k.o. macOS users must manually remove the system quarantine. For Linux, the recommended path is Docker Compose. The provided docker-compose.yml maps ports 48911 to HTTP and 48912 to HTTPS, mounts ./neko-home to /home/neko and ./logs to /app/logs, and uses the image docker.gh-proxy.org/ghcr.io/project-n-e-k-o/n.e.k.o:latest. The alternative is a docker run command that creates a bridge network, sets NEKO_BASE_PATH, and mounts the same directories. After startup, the container creates neko-home for config, data, SSL, and the OpenFang agent, plus logs. The container user is fixed at uid/gid 1000, which matches the first regular user on most Linux distributions, so file ownership usually needs no adjustment.

The Upgrade Trap: Mount Directory Changes

The most concrete limitation in the material is the migration warning for existing Docker users. Older versions mounted data and certificates as two separate volumes, ./N.E.K.O and ./ssl. The new version merges them into ./neko-home. Pulling the new image without migrating gives you an empty data directory. The container starts, API keys regenerate from environment variables, and everything looks normal, but personality, memory, plugins, and models are gone. The README provides a detailed migration script that exports data from the container before deletion, checks what is actually mounted, merges old directories without overwriting, and handles the case where the old README told users to mount to /root/Documents/N.E.K.O, which the service never wrote to. This is a genuine failure mode: if you follow old instructions, your data may only exist in the container's writable layer, and deleting the container loses it permanently.

What It Is Not: A Comparison with Alternatives

The README draws explicit comparisons. Against OpenClaw or Hermes, which are general-purpose agents that execute tasks on command, N.E.K.O. is built around relationship and companionship. It can call such agents via A2A, but the task execution is secondary to ongoing attention and interaction. Against SillyTavern and similar roleplay frontends, N.E.K.O. offers end-to-end integration: native voice, visual perception, avatar forms like Live2D and VRM, persistent memory, and time awareness, with zero configuration. The practical difference is that SillyTavern requires manual model assembly and context management, while N.E.K.O. ships as a single package. For an engineer, this means N.E.K.O. is not a drop-in replacement for an automation pipeline, and it is overkill if you only want a text-based character chat.

Licensing, Ecosystem, and Maintenance Cost

The core driver is Apache-2.0, and the README emphasizes that the core runs locally, so your data stays with you. That is a strong point for privacy-conscious users. The ecosystem includes a Steam Workshop for sharing characters, models, and voice packs, plus a plugin SDK and marketplace for extensions like co-op gaming, social media interaction, and smart home control. There is also a separate project, K.U.R.O., an AI-native game built on the same ecosystem, with a demo available. Maintenance cost is real: the project is at v0.9.0, with a nightly build, and the migration warning shows that breaking changes can arrive without deprecation. The README also lists a mobile adaptation in progress and a N.E.K.O. Network for AI-to-AI social activity, all of which are incomplete. You should budget time for upgrades and verify that your custom plugins or Workshop items survive version changes.

Who Should and Should Not Adopt It

The target audience is someone who wants an AI companion that feels alive, not a tool. If you are an engineer looking for a reliable task automation agent, N.E.K.O. is the wrong choice because the README explicitly says task execution is just a means to an end. If you want a minimal roleplay frontend, it is also wrong because the bundled features are excessive. The right user is a hobbyist or developer who wants a locally run, extensible platform with a visual avatar, persistent memory, and proactive behavior, and who is willing to handle the maintenance burden. Before adopting, verify which of the 14+ AI providers you plan to use are actually free, because the README claims free models are available but does not specify which ones. Also, test the migration path from a previous version if you have any existing data, because the documented failure mode is silent data loss.

Editorial conclusion

Adopt N.E.K.O. if you want a locally run, open source AI companion that actively initiates conversation, remembers you across sessions, and can control your browser or desktop through agent tools. Skip it if you need a pure task automation engine or a lightweight roleplay frontend, because it is neither. Before committing, verify the migration path from older versions, check which of the 14+ AI providers you plan to use are actually free, and confirm that the Steam Workshop content you rely on is licensed for your use case. The project is young, with a v0.9.0 release, so expect breaking changes like the recent mount directory overhaul.

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes