This project is the RENT A HAL AI Operating Realm
rentahal\webgui.py rentahal\main.py rentahal\static\script.js rentahal\templates\index.html
An Open-Source Speech-Enabled Web gui For Ollama
- with support for Llama, Llava, Stable Diffusion
- A I App platform with API interfaces
- out of the chute support for claude API and huggingface API
Support out of the box:
-
Ollama - native fastapi()
-
Llama - native fastapi()
-
Llava - nativefastapi()
-
Stable Diffusion via std API
-
CLAUDE via API
-
HuggingFaces via API
-
Runs on a three-node array:
-
Windows 10
-
nodes for:
-
stable diffusion
-
backend realm orchestrator
-
chat, vision (same node)
-
human interfaces:
-
web interface for chat, vision, imagine
-
speech input and putput
-
See demo:
-
API first platform
-
Easy to design and use lightweight powerful AI appls
-
demo with source code on youtube:
-
It's secure, on prem AI with unlimited backend AI worker nodes
-
that scale into a unified web-based AI solution and AI API platform to power your business.
-
AI radio hosts discuss MTOR:
https://www.amazon.com/dp/B0F6BDXZYH/
0:00 remember flipping on the TV and seeing 0:01 Captain Kirk just casually talk to the 0:05 computer what is the current status of 0:08 the USS Enterprise oh yeah absolutely 0:10 that sort of effortless interaction that 0:12 childhood dream right of a computer that 0:15 truly understood you your voice the why 0:17 behind your words uh our sources 0:20 actually drop us right into that feeling 0:22 gene Rodonberry wasn't just imagining a 0:24 gadget he was thinking about well a real 0:27 partner and that vision that futuristic 0:29 idea it connects directly to what we're 0:31 digging into today m exactly we've got 0:35 the MTO welcome to the realm document 0:37 plus some interesting perspectives from 0:39 uh AJL and Claude and they all point to 0:41 something potentially quite different 0:43 not just an upgrade but maybe a whole 0:45 new way to interact with tech that seems 0:47 to be the core idea yeah a fundamentally 0:49 new approach okay let's really get into 0:51 this new way the document calls MTOR the 0:54 realm beyond the kernel so for you the 0:57 learner our goal today is simple well 1:00 maybe not simple but clear okay fair our 1:03 goal is clear what is MTR really how 1:06 does it actually work and you know why 1:09 should you care there's so much noise 1:11 out there we want to cut through that 1:13 find those aha moments without getting 1:16 bogged down in like super deep technical 1:18 weeds think of this as your fast track a 1:21 shortcut to understanding something that 1:22 might be well pretty groundbreaking and 1:25 we have this interesting mix of sources 1:27 from the big picture philosophy down to 1:29 some quite specific details but the main 1:31 takeaway right off the bat is that MTOR 1:33 is just different from the OS you're 1:36 probably used to very different forget 1:37 your usual mental model of kernels 1:39 demons all that stuff so the big picture 1:42 M0 this multi-tronic operating realm 1:45 it's not built like a typical OS stack 1:47 right it's event driven it's stateless 1:50 we'll get into what that means and uh 1:52 speech is designed as a primary 1:54 interface and that client server model 1:55 we all know nope they call it peerto-rem 1:58 which is interesting okay so let's 1:59 unpack that what really makes m 2:01 different the document hammers this 2:03 point a realm not a stack yeah and 2:06 that's not just words it signals a 2:08 different structure entirely think about 2:10 traditional OS layers right each one 2:12 relies on the one below it uh-huh like 2:14 building blocks mtor tries to break that 2:17 the realm idea allows for more dynamic 2:20 sort of independent AI components it's 2:23 not just renaming it's rethinking how an 2:26 AI focused system is built more 2:28 flexibility more resilience maybe 2:30 especially with AI changing so fast 2:32 exactly that seems to be a core driver 2:34 and that leads right into this event 2:36 driven by default idea right so 2:38 everything inside MTOR communicates 2:41 using these structured JSON messages 2:43 everything like user commands AI 2:45 responses all of it all of it user 2:48 inputs worker replies system status 2:50 checks the works the document calls it a 2:53 giant programmable AI event router with 2:56 perfect visibility wow okay that pinks a 2:59 picture like a nervous system where you 3:00 can see every signal pretty much which 3:02 you know has big implications for 3:04 understanding what's going on for 3:05 debugging for monitoring and then the 3:06 interface speech is the interface back 3:08 to Star Trek mhm that 3:11 computer wake word isn't just nostalgia 3:14 it's central the vision is really 3:16 natural language interaction getting 3:18 closer to that partner idea not just a 3:20 tool executing commands that seems to be 3:23 the aim a more fluid intuitive dialogue 3:26 now this next bit might throw people 3:28 stateless by design yeah this one takes 3:31 a second to wrap your head around so 3:32 each AI query each request you send to 3:35 MTR it has to contain all the info 3:37 needed for that request right there's no 3:39 assumed memory or saved state between 3:42 interactions they even compare it to 3:43 CICS which uh might ring a bell for some 3:46 listeners okay so why do that seems less 3:48 efficient maybe well the argument is 3:50 scalability and resilience because each 3:52 request is self-contained if one fails 3:55 it doesn't mess up others uh okay makes 3:57 sense in a distributed system with lots 3:59 of things happening easier to manage 4:00 across different machines too that's the 4:02 idea simpler deployment less complexity 4:04 for managing state everywhere but uh 4:07 it's important yes while the processing 4:09 is stateless they do mention persistent 4:11 trace logs so there is a record kept 4:13 right so it remembers what happened just 4:15 doesn't rely on past context for the 4:17 next action exactly stateless in the 4:19 moment but with a history okay now brace 4:21 yourself for this list no kernel no 4:24 sysols no demons no file systems yeah 4:27 yeah it's like they wipe the OS slate 4:29 clean seriously but they didn't just 4:30 leave holes right they have replacements 4:32 they do so instead of demons you have 4:34 brokers instead of file systems it's 4:36 more about queries and replies instead 4:38 of sysols you have events it's a whole 4:42 different vocabulary a different way of 4:44 thinking about computation it really is 4:46 and it reflects how it manages resources 4:49 like no traditional file 4:51 hierarchy instead MTOR relies on the AI 4:54 workers themselves to hold or fetch 4:57 knowledge so you ask a question and it 4:59 queries the right AI model like 5:01 summoning info instead of finding a file 5:03 that's a good way to put it summoning 5:05 information which fits better with how 5:07 AI works you know constantly evolving 5:08 knowledge okay so we have this realm 5:10 events speech stateless missing the 5:14 usual OS bits how does it actually work 5:16 the component right let's break down the 5:18 main parts first the universal broker 5:21 the intent router they call it yeah it 5:23 takes your voice command figures out the 5:25 intent not just the words but what you 5:27 mean exactly then it checks which 5:29 workers are available sends out the 5:30 tasks gathers the results back and it 5:32 handles other stuff too like the token 5:34 economy we'll get to and failovers 5:37 it manages the wallets handles things if 5:40 a worker drops offline throttles 5:42 requests if needed it's the central 5:44 coordinator the AI orchestra conductor 5:47 that's a decent analogy yeah okay then 5:49 the AI workers these do the actual 5:51 thinking they do the heavy lifting the 5:52 AI processing and they can be anywhere 5:55 like on my machine or somewhere else 5:57 both local on your land or remote over 6:00 the internet they just need a way to 6:02 report their health say what they can do 6:04 like I run this image model and talk 6:07 JSON and the supported ones I saw a 6:09 llama stable diffusion yeah a llama 6:11 stable diffusion 1.5 elev for multimodal 6:14 stuff proxies for claude and open AI and 6:17 importantly you can plug in custom 6:19 Python agents too so it's extensible and 6:21 they just announce themselves pretty 6:23 much self-register and if they go 6:25 offline the system handles it gracefully 6:28 no big drama okay what about safe Q 6:30 sounds like it slows things down hey Not 6:33 necessarily it's more about fairness and 6:35 order it's described as a real time 6:37 intent buffer an intent buffer yeah so 6:40 it schedules tasks using async methods 6:43 handles them first in first out 6:44 generally prevents one user hogging 6:47 everything right and it does soft 6:48 throttling if things get overloaded plus 6:50 it gives you feedback like your request 6:53 is number three in the queue ah 6:54 transparency okay that makes sense keeps 6:56 things orderly and visible and the last 6:59 piece of this core puzzle health 7:00 monitoring this is the self-healing 7:02 aspect the workers constantly report 7:05 their status like what what model 7:07 they're running are they okay exactly 7:09 model status uptime latency even VRAM 7:13 usage mhm if a worker fails its health 7:16 check it gets benched automatically 7:18 blacklisted yeah taken out of rotation 7:21 and if it recovers later and starts 7:22 reporting healthy again it gets brought 7:24 back in gracefully readded the system is 7:27 constantly taking its own pulse that 7:28 seems pretty crucial for a distributed 7:30 system like this less manual babysitting 7:33 definitely it's designed for reliability 7:35 okay let's shift gears a bit let's talk 7:38 money or well tokens the $9,000 token ah 7:43 yes the economic layer this is a really 7:45 interesting part of the design so it's 7:47 not just about paying for things it is 7:50 but the goal seems broader it's about 7:52 aligning incentives for everyone 7:54 participating in the realm like giving 7:56 people a reason to run workers maybe 7:58 contribute models precisely imagine 8:00 researchers getting tokens for valuable 8:02 models or you getting tokens for letting 8:05 your GPU be used it's trying to build a 8:07 self-sustaining ecosystem that could be 8:09 huge for open source AI right funding 8:11 development potentially yeah it's a 8:12 forwardthinking approach to 8:13 sustainability and who manages this 8:15 token the Rendahal Foundation right and 8:18 their charter emphasizes things like 8:20 minimal founder allocation no special 8:23 privileges transparent transparency in 8:25 distribution and crucially that the 8:27 token's value is based on its utility 8:30 within the realm not just speculation 8:32 utilitydriven value okay so how does the 8:35 pricing actually work dayto-day they 8:38 have this dynamic pricing mechanism also 8:40 called rental it's not fixed like surge 8:43 pricing for AI kind of yeah it 8:46 fluctuates based on system demand 8:48 there's a base cost and then a dynamic 8:51 multiplier kicks in based on Q latency 8:53 how long the wait is so if the system's 8:55 really busy it costs more essentially 8:57 yes it's a way to manage demand and 8:59 allocate resources if you really need 9:01 something now you might pay more if you 9:03 can wait maybe it's cheaper later okay 9:05 that's clever and the token flow itself 9:08 pretty direct you the user make a 9:09 request your wallet gets debited $9,000 9:12 tokens and the person running the AI 9:14 worker that did the job their wallet 9:15 gets credited the same amount and the 9:17 whole thing is recorded on the 9:18 blockchain simple transparent creates 9:20 that direct incentive to contribute 9:23 compute power exactly fair compensation 9:25 for contributing resources okay stepping 9:27 back from the mechanics what about the 9:29 guiding philosophy the principles behind 9:32 MTOR chapter 7 lays out several key ones 9:35 first universal one realm designed to 9:38 run on pretty much any device anywhere 9:40 you have Python and a browser so 9:43 platform and hardware inclusive imagine 9:45 that you know for you the learner the 9:48 same AI system potentially running on 9:49 your phone your laptop maybe smart 9:52 devices that seamlessness that's the 9:54 goal breaking down those silos between 9:56 devices and OSS second principle open 9:59 yeah and this is a strong commitment 10:01 it's GPLV3 license which is a pretty 10:04 strict open- source license right 10:06 requires sharing modification it does 10:08 and they add this eternal openness 10:10 clause explicitly Prohibiting closed 10:12 source forks patents secrets they really 10:15 want it to stay open seems so fostering 10:17 collaboration preventing lock in third 10:20 respectful focus on privacy and the 10:22 user's purpose first pretty concise but 10:24 important especially now yeah privacy is 10:26 huge fourth decentralized no central 10:30 server dependency core to the whole idea 10:32 sounds like more resilient less prone to 10:34 censorship that's the argument for 10:36 decentralization yes distributes power 10:38 fifth is scalable right designed to run 10:41 locally on one machine but also scale up 10:43 to local networks the cloud potentially 10:45 a huge swarm so from one user up 10:49 to planetary scale that's ambitious very 10:52 ambitious but it speaks to the 10:54 architectural design and finally this 10:56 idea of built for the people by the 10:59 people yeah emphasizing distributed 11:01 ownership open participation and they 11:04 also loop back to statelessness here 11:06 highlighting it for concurrency and 11:08 reliability it paints a picture of a 11:10 very um democratized AI infrastructure 11:13 where individuals are empowered yeah so 11:15 okay this all sounds fascinating 11:17 potentially revolutionary but how does 11:20 someone actually start how do you the 11:22 learner get your hands dirty right 11:24 chapter 8 you are the SISOP now running 11:26 your own realm this is where the dream 11:28 meets the code so to speak empowering 11:30 people to actually run it themselves 11:32 exactly making it practical and the 11:34 steps are they complicated surprisingly 11:37 they seem pretty straightforward at 11:38 least to get started clone the repo from 11:41 GitHub install the requirements standard 11:43 developer stuff so far then you launch 11:45 the broker with a Python command connect 11:48 a GPU node if you have one with another 11:49 command like uicorn something yeah 11:51 something like unicorn main.applama 11:54 then you just open your web browser to 11:56 localhost 500 talk to a computer and 11:59 speak the words that's the idea lowering 12:02 the barrier to entry and this flips the 12:04 script right you're not just a user no 12:07 you become the builder the host the sis 12:09 of your own little piece of this realm a 12:12 contributor there's no big central 12:14 company you're logging into it's your 12:15 realm that sense of ownership big 12:18 difference from just using a service a 12:20 key part of the decentralized philosophy 12:21 but running your own thing setup 12:24 maintenance does it handle that itself 12:27 chapter 10 talks about 12:28 self-configuration and self-healing yeah 12:30 this part sounds pretty impressive a lot 12:32 of automation designed to make life 12:33 easier okay like automatic configuration 12:35 what does that mean it means when you 12:37 run it for the very first time it 12:39 basically sets itself up creates the 12:41 database it needs defines the worker 12:44 tables initializes your wallet starts 12:46 the web UI sends an initial message all 12:49 automatically apparently so zero config 12:51 deployment basically which is great for 12:53 getting started quickly removes a lot of 12:55 that initial setup headache definitely 12:57 lowers the barrier then there's this 12:59 config.in file right it autogenerates 13:01 this on the first run it's designed to 13:04 be human readable easy to understand and 13:07 self-documenting so you can go in there 13:09 and tweak things like where data is 13:11 stored which AI workers you prefer 13:13 exactly toggle modules like speech or 13:15 webcam use redirect storage path set 13:18 preferences and if you mess it up or 13:20 delete it it just makes a new default 13:21 one with comments explaining everything 13:23 that's what it says very user friendly 13:25 acts like a blueprint you can modify and 13:27 the database it sets that up too 13:30 database bootstrap yeah it autonomously 13:32 creates the schema tables for users 13:34 sessions transactions system metrics 13:36 using simple tools like SQLite and shelf 13:38 no complex database setup required from 13:40 the user nope it even populates the 13:43 initial CISUP details and gives you some 13:46 demo tokens to play with manages its own 13:48 persistence and safe fallbacks what if 13:51 things go wrong it tries to recover 13:53 gracefully if your wallet file gets 13:55 corrupted it might restore a backup if 13:58 the config file is broken it can revert 14:00 to defaults updates worker info 14:02 automatically if it's outdated seems so 14:05 the idea is even if you make a mistake 14:07 MT0 tries its best to recover and keep 14:10 running a safety net enhances robustness 14:12 and the self-healing loop it's 14:14 constantly checking its own 14:15 configuration if a setting is missing or 14:18 invalid it tries to fix it using default 14:20 values continuous monitoring and repair 14:22 keeps the realm stable over time that's 14:24 the goal adapting and maintaining 14:26 integrity and finally the watchdog 14:29 sounds serious yeah it's the guardian 14:31 yeah relentlessly checking things like 14:32 the Q processor worker health API 14:35 accessibility if something's 14:36 unresponsive it tries to restart the 14:38 component and notifies the CIS appu like 14:40 an automated admin catching problems 14:42 early wow okay a lot of built-in 14:45 resilience so you've got your realm 14:48 running what about connecting with 14:50 others federation in the grid right 14:52 chapter 11 this is where it scales 14:55 beyond just your own setup federation 14:57 lets different MTR instances connect and 14:59 share building a bigger distributed 15:01 network of intelligence that's the 15:03 vision allows realms to collaborate 15:05 share resources so I could potentially 15:08 send a task from my local realm to an AI 15:12 worker running in someone else's realm 15:14 maybe in the cloud or on their local 15:16 network exactly it talks about how easy 15:18 it is to connect workers using uicorn or 15:21 an automated fast API agent unlocks 15:23 resource sharing potential expands the 15:25 capabilities beyond just your own 15:27 hardware mhm and the sysop interface 15:30 gives you control over this so you can 15:32 see the remote workers how tasks are 15:33 being routed block certain nodes if 15:36 needed yeah provides oversight and 15:37 management tools for your federated 15:39 setup you're still in control of your 15:41 realm and the task routing isn't just 15:42 random the broker is smart about it it 15:44 tries to be roots tasks based on worker 15:47 load latency the specific model needed 15:50 finding the best fit finding the best 15:52 fit yeah and it autob blacklists workers 15:56 that are consistently failing or slow 15:59 optimizes performance okay makes sense 16:01 now for the builders the developers out 16:03 there what's the angle mt0 is called the 16:06 first OS for AI native applications 16:08 that's a bold claim but it highlights 16:11 the design intent it's built for AI not 16:14 just running AI on top of a traditional 16:16 OS recognizes AI's unique patterns so 16:19 it's designed to orchestrate AI work 16:21 that seems to be the core idea makes it 16:23 potentially very suitable for building 16:25 AI powered tools and it's API first mhm 16:28 chapter 14 emphasizes that core 16:30 endpoints like uh APA Epision a pageant 16:33 developers can interact via websockets 16:35 or standard REST API calls making it 16:37 easy to integrate MTOR into other apps 16:39 or build new things on top yeah 16:41 fostering that ecosystem of tools and it 16:43 plays well with others external APIs 16:45 local models seems designed for 16:46 flexibility mentions integrations with 16:48 claude hugging face plus local services 16:51 like a llama elvis stable diffusion 16:53 model agnostic giving developers choices 16:55 right leverage the best tool for the job 16:57 okay one last crucial area transparency 17:01 how do you see what's going on they 17:02 highlight the built-in debug console in 17:05 the web GUI not just a log file but a 17:07 live dashboard exactly showing active 17:10 connections workers query times Q status 17:12 token flow top tasks errors yeah all in 17:16 real time they call it unprecedented 17:17 transparency wow that's actually really 17:20 valuable for debugging understanding 17:22 performance building trust even 17:24 definitely seeing is believing right and 17:26 the main interface also shows Q status 17:28 worker health system load live yep 17:30 chapter 4 mentions that too real-time 17:32 visibility into the guts of the system 17:34 empowers users and sisops okay so 17:37 bringing this all home this deep dive it 17:39 really shows MT0 as a fundamentally 17:41 different beast yeah a different 17:43 approach to operating systems truly 17:45 built for the AI age voice interaction 17:48 decentralization openness at its core 17:50 and those key strengths browser native 17:52 model agnostic token governed extensible 17:55 multi-user ready it presents this uh 17:57 compelling vision for maybe a more 17:59 intuitive democratized way to access and 18:02 use AI could simplify the whole complex 18:05 landscape of models and infrastructure 18:07 potentially so for you the learner as 18:10 you digest all this maybe reflect on the 18:13 implications what could an open 18:15 decentralized AI realm mean how might it 18:19 change our relationship with technology 18:21 moving from just tools towards something 18:24 more like intelligent partners 18:26 definitely food for thought maybe even 18:27 worth checking out the source code the 18:28 docs it raises big questions about the 18:30 future of computing and the role of 18:32 these open community projects in shaping 18:34 it indeed so on that note we'll leave 18:37 you with those powerful words from the 18:39 M0 document itself say it with us 18:41 computer let the realm begin