Friday’s ZenOS-AI
A Modular, Context-Aware AI Home Automation Framework for Home Assistant
AI / LLM disclosure
ZenOS-AI and this post were developed with AI assistance. AI tools were used for drafting, refactoring, documentation, debugging support, and test workflow assistance.
The project is reviewed, tested, edited, and maintained by the ZenOS-AI maintainers. AI output is treated as draft work, not as an authority. Runtime AI behavior, privacy boundaries, Home Assistant permissions, and safety-related limitations are documented below.
ZenOS-AI may connect Home Assistant Assist, local or cloud LLMs, MCP-style tools, Home Assistant services, REST commands, and external household systems. Please review the privacy, permissions, and safety sections before exposing entities, scripts, REST-backed actions, locks, garage doors, alarms, cameras, or other sensitive controls.
This is the support, release, and announcement thread for ZenOS-AI!
(the underlying technology for ‘Friday’s Party’)
Current
Stable: 2026.7.1
Beta: 2026.8.0 (TBD)
ZenOS-AI is a third-party project for Home Assistant. It is maintained by the ZenOS-AI project maintainers. It is not part of Home Assistant Core and is not maintained by Nabu Casa or the Home Assistant project.
Please use this thread for:
- Release announcements
- Upgrade notes
- Installation help
- Configuration questions
- Bug reports
- Compatibility reports
- Documentation corrections
- Questions about supported ZenOS-AI capabilities
For broader discussion, demos, experiments, architecture ideas, screenshots, weird discoveries, and “what are people building with this?” conversation, please continue using the long-running Friday’s Party thread:
Front door to the party. < I’m not going to apologize, enter at your own risk.
If you have followed Friday’s Party, you may have seen me banging my head against this idea in public for the last 18 months. That thread is now over 600 posts old today. 
Yay.
It is fun, messy, useful, and very much still alive.
This thread is for the people who do not want to read the whole saga.
If your style is more “give me the release notes, tell me what it does, show me how to install it, and I’ll ask for help if I need it,” this thread is for you.
Think of this thread as the support desk.
Friday’s Party is still the campfire.
The short version
Everyone is suddenly talking about “AI agents” in the home.
ZenOS-AI is my attempt to build one that is useful, inspectable, and leashed. You want the ‘why’ go read the party… 
It lets Home Assistant behave more like a capable ‘household assistant’ while keeping control inside Home Assistant’s own automation model. The assistant can reason over the context you provide, use the tools you expose, call the scripts you allow, and interact with external systems through Home Assistant-native patterns.
That gets stupid powerful fast.
Sure, it can brief the house. …So can the default HA AI Task demo. A ZenOS Agent can know where your STUFF is. (That’s a different animal entirely) Knowing CONTEXT, where a thing is how it’s used what it connects to - unlocks things like:
- It can connect inventory, documents, service requests, rooms, media, projects, people, and money into something the assistant can actually reason over.
- It can help guide a guest through a configured emergency routine without exposing private household details.
- It can work with an ITIL-style service desk system because Home Assistant can call REST endpoints, and ZenOS-AI can wrap those actions in safer assistant-facing workflows.
- It can play Zork in your actual house. Totally not as a cranky AI who DIDN’T get selected for this season of Dungeon Crawler World. He MIGHT even be a little spiteful about being STUCK in an IBM AT… (please don’t sue me Matt. Everyone go read AND LISTEN TO DCC wherever you buy your audiobooks!)
That last one sounds silly until you realize what it proves: state, memory, room topology, inventory, navigation, multi-step interaction, and safe simulation.
The wow is real.
The leash is real too.
What makes ZenOS-AI different?
ZenOS-AI is not a Python integration, an add-on full of custom services, or an external MCP server bolted onto Home Assistant.
This repository contains zero executable Python code.
ZenOS-AI exists ENTIRELY inside the Home Assistant agent box. It can only act through Home Assistant constructs that you configure and expose.
It is built from:
- Home Assistant packages
- Jinja templates
- Scripts
- Automations
- Helpers
- Scenes
- Intents
- Assist-exposed tools
- Home Assistant services
- REST commands
- Native Home Assistant configuration patterns
Most AI smart-home projects start by standing up a separate agent runtime and wiring that runtime back into the house, usually through MCP. “Just give me an API Key - we got you boo…” (Yeah, the alarm bells are going off in my head too. They should. I AM NOT saying all of these are bad, I AM saying you need to do your research first. but I digress…)
INSTEAD, ZenOS-AI starts from inside Home Assistant. It asks how far Home Assistant itself can go when its native building blocks are organized for AI-assisted workflows.
The important caveat is REST Command.
REST Command is a native Home Assistant construct. ZenOS-AI leans on it heavily. REST commands let Home Assistant call external HTTP endpoints from scripts, automations, templates, and services configured by the user.
That means ZenOS-AI can reach outside Home Assistant responsibly when the user deliberately configures those calls, while keeping the action path inside Home Assistant’s own tooling model.
In plain English: 
ZenOS-AI can talk to external systems, create tickets, query inventory, update records, retrieve documents, log expenses, or trigger workflows if you configure those capabilities through Home Assistant.
The assistant can only do what Home Assistant tools, intents, scripts, automations, templates, helpers, scenes, REST commands, and services allow it to do.
What ZenOS-AI is
ZenOS-AI is a Home Assistant-native AI framework for building private, context-aware, agentic smart-home assistants using Home Assistant’s own tools and automation primitives.
It adds structured memory, tool manifests, knowledge routing, room topology, summarization, identity-aware context, and higher-level household workflows on top of Home Assistant.
It is designed to help Home Assistant Assist and compatible LLM backends work with structured home context, rooms, inventory, media, reminders, documents, household routines, service requests, and external systems reached through user-configured Home Assistant services or REST commands.
The goal is simple:
Give Home Assistant better context, safer tools, and more useful AI behavior while keeping the user in control of what is exposed, what can be acted on, and where data goes.
Current release
Stable: 2026.7.1 “Neo+”
Neo is the release where ZenOS-AI turns from “a collection of useful tools” into a system where the tools know themselves, talk to each other, and reason from live structured memory.
Highlights include:
- CabCeption and LiveDrawers: structured memory becomes a graph, not a flat key-value store.
- Tool Manifest: tools self-describe, including what they consume, return, require, and risk.
- Lens Bus: generic knowledge requests route to registered providers like documents, wiki, service desk, media, and finance.
- Room Manager: spatial topology, room context, exits, safety equipment, inventory, tasks, tickets, and media context.
- ZenZork: a live text-adventure engine running on actual Room Manager topology.
- New plugin surfaces for service desk, documents, wiki, CRM, finance, and kitchen workflows.
PLUS
-
KF5 — Self-Registering Tools:
bootstrap_kfc— tools declare their own KFC dojo-drawer contract viamode=kfc_manifest;zen_dojotools_manifestdiscovers and mounts them automatically, no Scribe authoring step needed. Adopted by Room Manager, AlertManager, Camera, Security Manager, and SystemTools.- FileCabinet mode-scoped whitelist —
tool:modecallout entries let a live drawer reach exactly one mode on a tool (e.g. a read-only self-description), never a write path.
-
Finance & Inventory:
- Firefly III codex tier —
zen_codex_finance_depreciation(asset depreciation: SL/DDB schedules, business-use splits, disposal) andzen_codex_finance_cogs(auto-posts COGS to Firefly from Grocy stock consumption). - Grocy v5.3.1 — fixed
stock_entry_updatesilently sending empty payloads on entry-id-only calls. New: Perishable Storage Coaching, COGS Coaching, battery-stock tracking, tax/depreciable-asset userfields.
- Firefly III codex tier —
-
New Lens Bus Provider:
- Battery Notes (
zen_stack_battery) — first ZenOS stack provider backed by a HACS integration (ha-battery-notes) rather than a self-hosted service. Battery health by area, cross-referenced with Grocy stock.
- Battery Notes (
-
Other:
- Postman
direct_dispatch— authority-stack bypass mode, the migration path for the retirednotification_routerscript. - ZenZork v1.7.0 —
llm_narrationtoggle for live LLM-generated room narration, template fallback.
- Postman
What can my agent ACTUALLY do?
Home Assistant already provides the primitives: Assist, scripts, automations, templates, helpers, services, entities, labels, scenes, intents, and REST Command.
ZenOS-AI uses those primitives to build higher-level assistant experiences.
1. Brief the house
Ask for the state of the home in human spatially and contextually anchored terms.
Friday can assemble a situational view from rooms, alerts, utilities, media, inventory, tasks, calendars, service tickets, and system summaries.
- “What needs attention right now?” (Yes, with the summary pipeline active, she can.)
- “What changed since this morning?”
- Wake “You left the trash out again, Nathan…” < yeah, she does that too.
A smart home should do more than report raw entity states. It should help you understand what matters. AI is more than Rainbows and unicorn stories… If you’re going to burn the power and compute, make it useful.
2. Know where things are
With inventory configured, the agent can help locate supplies, parts, tools, equipment, books, games, pantry items, and project materials.
- “Where are the sublimation supplies?”
- “What spare parts do we have for the hot tub?”
- “Add this printer to inventory and mark it as converted for sublimation ink.”
^^^ and when you use said agents to help write the team’s intro, they use your project list, apparently…
Most smart homes know devices. ZenOS-AI helps the home know stuff.
3. Run household maintenance like operations
With the service desk plugin configured, the agent can create, search, update, triage, link, and close work requests. Maybe a leaking sink, a broken blind, a hot tub issue, or a project task can become a tracked operational record instead of a forgotten text message.
- “Can you report that the sink is leaking?”
Creates a service ticket, not a whole home announcement. The difference is CONTEXT.)
- “Find the notes from the last AC issue. Who did we call?”
Context linking allows indexed search to start in the room or concept anchor HVAC is a concept not an entity.
- "Did we actually replace the filter in the lawn tractor last time?”
Grounded in an ERP-capable inventory that you don’t need to be a quartermaster to operate - LLM > greek to me translation turns the lawn tractor into a ‘managed machine’ that exists in inventory anchored to the garage.
The house starts to feel like a tiny facility with it’s own facilities manager who just ‘knows.’ ZenOS-AI lets Home Assistant participate in the operations layer.
4. Understand rooms as places
Room Manager gives your agents spatial context.
Rooms can have portals, adjacency, exits, windows, safety equipment, utility notes, inventory, tasks, tickets, wiki pages, media state, and emergency guidance.
- “What room am I in?”
- “What exits are available from here?”
- “What tickets, tasks, inventory, and media are tied to the kitchen?”
- “Where is my Super-Suit?”
Rooms are not just labels on entities. They are places with context, risks, contents, and relationships. And the whole thing exists inside existing HA frameworks just looked at from a ‘certain point of view.’
5. Coordinate real home systems
ZenOS-AI includes higher-level managers for real household systems.
- AutoVac for vacuum scheduling, readiness, consumables, wear monitoring, and post-run analysis.
- SpaMaster for hot tub status, scenes, chemistry, post-soak logs, and supplies.
- Media Manager for whole-home AV discovery, now-playing context, source preferences, and playback routing.
- Security Manager for alarm state, zones, cameras, policies, and spatial handoff.
- Plant Manager for power, water, gas, HVAC, mechanical systems, circuits, and utility context.
- More coming. All while trying to be judicious with your tool count.
- “Check the spa and tell me what needs attention.”
- “Is the vacuum ready to run?”
- “What is playing in the living room?”
- “Show me the house utility overview.”
6. Find knowledge across systems
ZenOS-AI’s Library and Lens Bus route knowledge requests to registered providers.
That can include wiki pages, Paperless documents, service desk tickets, media evidence, catalog records, and other configured knowledge stacks.
- “Find the manual for the hot tub.”
- “Search the house notes for sublimation ink.”
- “Find everything we know about the printer conversion.”
A useful assistant should know where to look without making the user remember the exact system, file name, tag, or API.
7. Connect people, vendors, projects, money, and work
Optional plugins can connect Home Assistant workflows to CRM, finance, and project tracking.
That means the agent can help relate household work to contacts, companies, receipts, assets, budgets, bills, transactions, projects, service requests, and follow-ups.
- “Who is the vendor tied to this repair?”
- “Attach this receipt to the deck project.”
- “What did we spend on the printer conversion?”
- “Mark this tool as a business asset.”
- “Create a follow-up task for that contact.”
Because, some homes are also workshops, studios, rentals, labs, small businesses, or mini family operations centers.
8. Guide guests and emergencies safely
ZenOS-AI can support configured guest-safe and emergency guidance workflows through Home Assistant Assist, notifications, speakers, displays, scripts, room context, and automations.
This is not a replacement for alarms, emergency services, fire safety systems, locks, cameras, or professional life-safety equipment.
It can help communicate instructions when the right Home Assistant routines are already configured.
(All below would require manual configuration, we do not configure triggers by default, these are absolutely possible using known patterns and standard HA automations - even the new triggers!)
- “A guest is here and the smoke alarm routine is active. Tell them what to do.”
- “Announce the safest exit route.”
- “Notify the household that a guest needs assistance.”
- “Explain what is happening without exposing private household details.”
Because a smart home should be useful to people who do not know how the house works, especially when something is wrong.
9. Play Zork in your actual house
Yes, really. 
ZenZork is a text adventure engine that uses live Room Manager topology. It can navigate rooms, track facing, resolve exits, carry inventory, examine spaces, run quests, and produce a post-game Room Manager quality report. You need a way to check your work. Might as well be interesting.
Examples:
- “Start Zork mode.”
- “Look.”
- “Go north.”
- “Take the flashlight.”
- “Start a quest to explore every room.”
If the agent can navigate a live model of the house as a text adventure, she can also support guided setup, troubleshooting, walkthroughs, inventory capture, and mock emergency flows without touching sensitive devices.
The relentless mocking of the dungeon when you failed to find the sock the 3rd time? You’re on your own buddy. I cannot guarantee you will not be eaten by grue nor certain goblin explosives do not show up on your path - stuff 'em in inventory quick! I dunno, you figure out what to do with unexploded inventory that’s your problem. Ask Carl. (You did see me say go read the book right?)
What ZenOS-AI does not promise
ZenOS-AI is powerful, but it is not magic and should not be treated as a safety-certified control system.
ZenOS-AI does not promise:
- Perfect AI answers
- Perfect device control
- Correct behavior from every LLM backend
- Compatibility with every custom integration
- Safe automatic control of sensitive devices without careful configuration
- That every optional plugin is available in every install
- That cloud models keep data local
- That local models are automatically private if the surrounding system is misconfigured
ZenOS-AI also does not bypass Home Assistant.
It cannot execute arbitrary Python from this repository, install packages, access the host system, or perform actions outside the Home Assistant tools, intents, scripts, automations, templates, helpers, scenes, REST commands, and services you configure.
Installation and documentation
REQUIRES A VALID CONVERSATION AGENT ALREADY SETUP - SEE HA documentation about installing a conversation agent of your choice.
ZenOS-AI installs as a Home Assistant package collection. Full installation steps are in the repo.
-
Install guide: zenos-ai/zenos_ai/docs/getting_started/install.md at main · nathan-curtis/zenos-ai
-
Documentation hub: zenos-ai/zenos_ai/docs/readme.md at main · nathan-curtis/zenos-ai
-
The Party: Friday's Party: Creating a Private, Agentic AI using Voice Assistant tools
Basic install shape:
- Copy the ZenOS-AI package directory into your Home Assistant
packages/folder. - Copy the ZenOS-AI custom templates into your
custom_templates/folder. - Enable Home Assistant packages in
configuration.yamlif not already setup for packages. - Setup API keys for external services or dummy keys as desired.
- Paste the one-line conversation agent prompt template into your preferred - configured conversation agent.
- Set the ZenOS conversation agent helper to your conversation agent entity.
- Restart Home Assistant.
- Check ZenOS health sensors and run first-run setup.
See the install guide for the full walkthrough.
Optional plugins are disabled by default and enabled only if you need them.
Getting help
When asking for help, please include:
Home Assistant version:
ZenOS-AI version:
Install method:
Conversation agent:
LLM backend:
Local or cloud model:
Relevant ZenOS-AI component or plugin:
Relevant REST commands, if any:
What you expected:
What happened:
Relevant logs:
Steps to reproduce:
Please redact:
- API keys
- Tokens
- Webhook URLs
- REST endpoint secrets
- Personal names
- Street addresses
- GPS locations
- Private entity names
- Calendar details
- Message contents
- Anything else you would not want public
Please do not paste unverified AI-generated troubleshooting answers as if they are known fixes. If AI helped you investigate something, say so and include the actual logs, configuration, or test results.
Bug reports
For bugs, please include:
- ZenOS-AI version
- Home Assistant version
- Install method
- Relevant component or plugin
- Relevant configuration
- Relevant REST command configuration, if involved
- Error logs
- Steps to reproduce
- Expected behavior
- Actual behavior
- Whether the issue started after an upgrade
Bugs should be filed here:
Issues · nathan-curtis/zenos-ai
This thread can still be used to ask whether something is likely a bug before opening an issue.
Release announcements
Release posts in this thread will generally include:
- Version number
- Release date
- Compatibility notes
- Breaking changes
- New features
- Bug fixes
- Migration steps
- Known issues
- Documentation links
Current capability areas
ZenOS-AI 2026.7 includes or documents these major areas:
| Area | What it adds |
|---|---|
| Room Manager | Spatial topology, room context, emergency snapshots, exits, inventory, tasks, tickets, media, wiki, utility index |
| FileCabinet | Structured memory, drawers, nested graph storage, VirtualDrawers, LiveDrawers |
| Tool Manifest | Self-describing tools, health, risk class, modes, inputs, outputs, routing metadata |
| Library and Lens Bus | Knowledge routing across documents, wiki, tickets, media, catalog, and other providers |
| Grocy Inventory | Governed inventory, stock, locations, shopping, chores, catalog, assets, consumables |
| AutoVac | Autonomous vacuum scheduling, room election, readiness, consumables, wear monitoring |
| SpaMaster | Hot tub control, scenes, chemistry, logs, consumables |
| Media Manager | Whole-home AV, now-playing context, source prefs, playback hints, media evidence |
| Security Manager | Alarm panel, zones, cameras, security lens handoff, alert policy |
| Plant Manager | Power, water, gas, HVAC, mechanical systems, circuits, utility registry |
| Zammad Radar | Service desk tickets, work requests, triage, anchors, queues, service history |
| Twenty Rolodex | Contacts, companies, vendors, stays, appointments, reminders, tasks, HA person links |
| Firefly III | Accounts, transactions, budgets, bills, savings goals, liabilities, finance snapshots |
| Project Tracker | Project manifests, receipt logging, Grocy and Firefly linkage, reconciliation |
| ZenZork | Text adventure engine on live Room Manager topology |
Some areas are optional. Some require external systems. Some are power-user territory. Release notes will call out what is stable, experimental, or scaffolded.
Closing note
ZenOS-AI is for people who want Home Assistant to become more context-aware, more useful, and more personal while keeping control of their home.
The project asks a fun question:
How far can Home Assistant go as an AI assistant platform when you refuse to hide the magic in a black-box Python integration?
So far: MUCH farther than expected.
Use this thread for support and releases.
Use Friday’s Party for the bigger conversation, experiments, screenshots, ideas, and community builds.
Party on, Friday.
yeah pretty cool, right?
youve done it