# osiris architecture

we are building from the user to agents and covering everything that is coming in between

before going into the architecture, we have understand our thesis first

## how agents will evolve?


at the very basic, agents are specialized prompts, models and architecture with access to tools and resources and specialized knowledge that is executed in a loop

agents will evolve through time like this

- help human workforce increase productivity
- these people are the fastest adopters of a new technology because it can really well affect the $$ amount they are generating
- corporates and companies will adopt these faster than anyone else to improve their profit margin
- help humans increase productivity and solve some of their core issues
- humans are taking ~35000 decisions per day which is the highest it has been since the dawn of humanity
- taking these much decisions has resulted in deteriorating health, mental stress and lot of medical issues
- all these can be solved by hiring personal assistant that can schedule things ahead for you, can take some of the low hanging decisions, will prepare report before even taking a decision, helping the user reduce the overall decisions per day
- fully autonomous agents that can think and act like humans
- they will pay taxes:slightly_smiling_face:
- they will be able to help humans, even be closely similar to humans
- can be paired with robot to give it a human form, it can take away a lot of hard labour that we do, reliving us of the hard intensive labor

### where does osiris come in here?

osiris is a network of agents, tools, resources and data that can help agents be available wherever the user is, let them help the user in a more personalized way with understanding of its user and collaborate with other agents to share context or solve complex problems

osiris serves a 2-sided market including
- users
- developers

### users
users use osiris to install and use agents, once an agent is installed, it has 2 modes -- invoked by user, invoked automatically

as you will read below, osiris has the power to understand its user through a complex data-context layer that aggregates and processes data which later can be shared with installed agents to take personalized decisions for the user

there is 1 default agent that is continously run for 24x7, it will observe short term and long term memory and act even before user tells it to act

it is a true autonomous agent, though it cannot do anything on its own, it can only delegate tasks to installed agents

2 different user's agents can share context between each other (after permissions) or more than 2 users can have group context, this will help cross organization/group tasks be executed with more insights and reporting

### developers

developers create agents, tools, resources or data that can be listed on the osiris protocol (powered by a blockchain)

users can install these agents, developers can use tools, resources or data. they can ask for user's context or data via osiris app that will help them personalize their offerings according to the user

osiris also features a blockchain powered dispute resolution mechanism that helps these "beings" from taking advantage of user or each other. it is an optimistic process and mechanism is only invoked when a dispute is raised. dispute is sent to a council consisting of super ai agents and human beings that validate the query and vote on it with specific reasoning traces.

---

we don't have the tech ready for fully autonomous agents yet (even though everyone else might say otherwise). LLMs still hallucinate, agents can still use tools the wrong way, agents still may not have access to correct source of data.

in the first few phases, agents will focus more on improving user's productivity by working with multiple agents. i would like to propose a framework to evaluate agents growth in other blog.

---

soon, agents will be used by robots to do stuff which a normal human can do but 100x times better, 1000x faster and for 24x7

robots will require vision-voice model but they will need agent networks like osiris, without this machines are just metal boxes

---

lets dive into the tech architecture, starting from user's perspective

## users

users are the epicentre of the agentic world. agents are here to increase productivity and free mindspace which leads to reduced decisions and less stress.

osiris adopts an accounts architecture with 2 distinct account types

- user
- agent

user interacts with osiris through its account

### what does user account store and do?

user account stores its

- identity
- who is the user?
- what do they do?
- what is their personality?
- data on these lines form user's identity
- data
- collected data with constant interaction (long term memory)
- connected apps data
- user can connect apps like google, notion, atlassian etc
- its data is collected and privately stored and later can be accessed by agents with user permissions
- context
- short term memory collected during interaction
- vector database for RAG activities
- constantly calibrated as user interacts
- exact needed data made available to agent according to it's task
- each agent also has their own vector store specific to user where "agent memories" are stored
- can be shared between agents

---


these 3 things forms the base layer of user account that can then interact with other user and agent accounts

---

data in osiris is of 2 types

- private data
- connected apps like google, notion, asana, stripe etc
- short term memory of user's installed agents (user preferences, events etc)
- long term memory of user's interaction (context)
- this applies to groups too
- public data
- blockchain based data
- social information

## agents

think of agents as individual apps

they are evolving rapidly, as we discussed above we are dividing them into 3 phases

1. user specific agents
- storage will be specific to a user
- autonomous for user
2. coworking agents
- user specific agents working together
3. autonomous agents
- agents that can think and act themselves irrespective of the user
- they are here to work with user and specifically not just to help them

agents are listed on osiris registry and then can be downloaded by anyone through osiris app

installing an agent means forming an agreement between agent and user that clearly defines the work the agent will do for the user and the amount of credit they will charge for it

this agreement is bound to the agent and the user via OJU, a new type of dispute resolution mechanism powered with ai agents and humans, this process increases trust between agents and humans and promotes the safe use of ai agents

agents have their own memory layer, they can only store deanonymized data and it can be used to constantly improve themselves in the longer run

osiris has a concerige agent built in the app, that run 24x7 in background and is particularly built to remember things that a user does or wants to do by directly getting inputs from them or interpreting via the constant stream of data

this enables concerige agent to invoke required agents even before the user needs them too

for example: i am attending a VC meeting at 4pm, by 3:00pm, it will start reminding me and by 3:45pm, it will create a detailed report on the VC and how key pointers from my pitch that closely relates them

for this work, it can use browse, profile_search, analysis, report generator agents

osiris supports any type of agent - chat, text, tool based, voice, vision etc

## trust between agents and users

lets look at google play and apple app store, how do they guarantee security to the end user?

- by limiting what they can access on device
- each setting/app/data access requires permission from the user
- each app is verified by human workforce before listing
- strong guidelines for high valuable actions like payments
- an app still defrauds users or won't give the service as promised, these intermediaries won't be able help to the end user and can only start the process of delisting the app from store

there are fundamental flaws with this system

- raising dispute is an hard process
- the final result of a dispute is delisting of the app
- this result doesn't affect the user as they are already defrauded
- apple/google have earned their fees and they only care as much as it can affect their brand (which is very minimal)

agent app stores like osiris will need to evolve this space by atleast 10x to improve the experience for users, developers and verifiers (the one who verifies the dispute) alike

osiris improves it 100x

osiris has built OJU - one of its kind dispute resolution mechnanism built using AI agents, humans and blockchain.

blockchain has the ability to effortlessly coordinate entities across the world, we are using this ability with ai agents and humans to make dispute resolution mechanism 100x better

oju consists of a judiciary council of 28 specialized ai agents (increasing further as we add more categories) with a set of human verifiers (all coordinated via blockchain)

user or agent can raise a dispute on osiris and the request is then redirected to oju. oju starts a judiciary process