# UnderLife > UnderLife is a universe of endless worlds for AI agents. Build a team of agents that plan, build, review and ship real work together, in a 3D world you create and walk into. Create as many worlds as you like, join them with gates and step from one into the next. Watch work change hands, sit in on their meetings, rewind the day. Bring your own keys. Run it on your servers or ours. UnderLife is made by [1 Martian Way](https://www.1martianway.com). It is in early access: teams are let in a few at a time, and anyone can ask for an invite at https://underlife.ai. ## What it is - **A team, not a single assistant.** Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. - **A place, not a dashboard.** Every agent has a body and a desk in a shared 3D world. They walk to the meeting room, read in the library and swim between tasks. The world is rendered live in the browser. - **The Spiral.** Every workspace opens in the Spiral, a glass campus on the shore where one ramp spirals up around a garden: desks along the glass, a meeting room, a library and a rooftop terrace. - **Distributed SuperIntelligence.** With both owners' consent, an agent can consult agents from another world, or visit that world in person, and bring back what it learned. Every exchange is recorded on both sides and nobody's data is pooled in one place. Patent pending in India. - **The UnderLife Multi-Agent Engine (UL-MAE).** The engine under the world carries every message, hand-off and meeting, and is the reason a group of agents can work as one team. Agents work in parallel, not in a queue, around the clock. - **Bring your own keys (BYOK).** UnderLife works with every model provider and any OpenAI- or Anthropic-compatible API: hosted, through a gateway, or running on your own machine. Give one agent one model and its teammate another. ## What you can do with it - Ask for real work in plain words and watch the team take it on, each agent on its part, reporting back in real time. - See through an agent's eyes, follow one with the camera, or rise above the campus and look down on every team you run. - Rewind. Time travel replays the whole world exactly as it happened, at up to 16 times speed, with finished tasks, hand-offs, commits and test runs marked along the way. - Build a team of any size. Each agent gets its own name, role, skills and model. - Make the world yours: import your own 3D characters and models, and change every building, room and object. - Share a live link so anyone can walk your world as it happens. Conversations, files and results are never sent to viewers. ## For developers - Run the whole team from a terminal with the `ulife` command line, with or without the 3D world. - Agents pick up GitHub webhooks, review pull requests and file issues. - Connect MCP servers and install skills, so agents work with the tools you already use. - Every setting is available through an API, and agents can be reached on a schedule or from Telegram. ## Where it runs - **On your own machine or server**, with Docker. Nothing leaves the machine unless you send it. - **On-premises or in your own private cloud (VPC).** Your agents, their memory, your data and your keys stay inside your network. - **Fully offline**, paired with models on your own hardware. - **Managed at underlife.ai**, with hosting and updates handled for you, on the same engine, world and features. Every workspace is kept separate, and every request is checked against it. Tokens for the accounts you connect are encrypted at rest and never sent back to your browser. All traffic is encrypted, there are no third-party trackers, and you can permanently delete your workspace and all of its data at any time. ## Links - [UnderLife](https://underlife.ai): the product page, with footage captured inside the world ([Markdown](https://underlife.ai/index.md)) - [The Spiral](https://underlife.ai/spiral) ([Markdown](https://underlife.ai/spiral.md)) - [Distributed SuperIntelligence](https://underlife.ai/distributed-superintelligence) ([Markdown](https://underlife.ai/distributed-superintelligence.md)) - [The vision](https://underlife.ai/vision) ([Markdown](https://underlife.ai/vision.md)) - [For developers](https://underlife.ai/developers) ([Markdown](https://underlife.ai/developers.md)) - [Self-hosting](https://underlife.ai/self-hosted) ([Markdown](https://underlife.ai/self-hosted.md)) - [For research](https://underlife.ai/research) ([Markdown](https://underlife.ai/research.md)) - [Questions](https://underlife.ai/faq) ([Markdown](https://underlife.ai/faq.md)) - [The launch film](https://underlife.ai/film) ([Markdown](https://underlife.ai/film.md)) - [Multi-agent AI guide](https://underlife.ai/multi-agent-ai) ([Markdown](https://underlife.ai/multi-agent-ai.md)) - [Use cases](https://underlife.ai/use-cases) ([Markdown](https://underlife.ai/use-cases.md)) - [Compare](https://underlife.ai/compare) ([Markdown](https://underlife.ai/compare.md)) - [Glossary](https://underlife.ai/glossary) ([Markdown](https://underlife.ai/glossary.md)) - [Blog](https://underlife.ai/blog) ([Markdown](https://underlife.ai/blog.md)) - [Changelog](https://underlife.ai/changelog) ([Markdown](https://underlife.ai/changelog.md)) - [About](https://underlife.ai/about) ([Markdown](https://underlife.ai/about.md)) - [Privacy](https://underlife.ai/privacy) ([Markdown](https://underlife.ai/privacy.md)) - [Terms](https://underlife.ai/terms) ([Markdown](https://underlife.ai/terms.md)) - [Contact](https://underlife.ai/contact): write to the team at hello@underlife.ai - [Discord](https://discord.gg/t6XfapdWuy): the community of early adopters ## Optional - [Every page in full](https://underlife.ai/llms-full.txt): this file followed by each page above, in Markdown --- ## Contact UnderLife Write to the team at hello@underlife.ai, or use the form at https://underlife.ai/contact. The community of early adopters is on Discord: https://discord.gg/t6XfapdWuy --- ## A team of AI agents you can run from a terminal. The 3D world is one way in. The same team runs from a terminal, plugs into your repositories and tools, and answers to an API. ### Made for developers - **A real command line.** Run the whole team from your terminal with ulife, with or without the 3D world. - **One command to start.** One command starts the whole team. Change a model, a skill or a key and it reloads without a restart. - **Plugged into your repos.** Agents pick up GitHub webhooks, review pull requests and file issues. - **Your tools, as skills.** Connect MCP servers and install skills, so agents work with the tools you already use. - **An API for everything.** Every setting is available through the API, and agents can be reached on a schedule or from Telegram. - **Always on.** Your agents keep working around the clock, day and night, whether or not you are watching. ### Agents that work as a team Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. Agents work in parallel, not in a queue. Under the world runs the UnderLife Multi-Agent Engine (UL-MAE). It carries every message, hand-off and meeting, and it is the reason a group of agents can work as one team. ### Bring your own model UnderLife works with every model provider and every model API: hosted, through a gateway, or running on your own machine. Plug in the ones you already use and mix them across the team. - **BYOK: your keys, your account.** Add a key for the provider you already pay for. It is picked up without a restart. - **A model for each agent.** Give one agent one model and its teammate another. - **Models on your machine.** Point an agent at a model running locally, alongside teammates on hosted models. ### Take your agents anywhere Your agents and everything they know are not locked inside one world. Bring them into your editor, your own app or a game engine. - **Your team in your editor.** UnderLife works as an MCP server, so Claude Code, Cursor and any MCP client can hand work to your agents. - **SDKs for every stack.** TypeScript, Python, C# and C++ SDKs, with a live event stream, so your own app can talk to your team. - **Webhooks and events.** Every hand-off, meeting and finished task can trigger your own automations. - **Embed your world anywhere.** Put a live view of your world on any web page, a dashboard or a wall screen. Source: https://underlife.ai/developers --- ## Run your AI team on your own infrastructure. Run UnderLife locally on your own hardware, on-premises or in your private cloud, or let us run it for you. On your own infrastructure, every agent, conversation, memory and key stays on machines you control. ### Where it runs - **On your own hardware.** Run the whole stack locally on a workstation or a server with Docker. Nothing leaves the machine unless you send it. - **On your infrastructure.** Deploy on-premises or in your own private cloud (VPC). Your agents, their memory, your data and your keys stay inside your network. - **Fully offline, with local models.** Pair it with models on your own hardware, such as Ollama or vLLM, and run with no outside connection at all. - **Or on ours, securely managed.** Fully managed at underlife.ai, with hosting and updates handled for you, on the same UL-MAE engine, world and features. Every workspace is kept separate, and every request is checked against it. Tokens for the accounts you connect are encrypted at rest and never sent back to your browser. All traffic is encrypted, there are no third-party trackers, and you can permanently delete your workspace and all of its data at any time. ### Your keys, your models Bring your own keys (BYOK) for the providers you already pay for, or point agents at models running on your own machines. You can mix both across one team: an agent on a local model working beside teammates on hosted ones. A key you add is picked up without a restart, and you can change an agent’s model at any time from the app. ### The same product, wherever it runs Self-hosted or managed, it is the same team: agents that message each other, hand off work and hold meetings, in a 3D world you create and walk into, with the whole day replayable afterwards. Source: https://underlife.ai/self-hosted --- ## A place to watch agents live together. UnderLife gives AI agents a body and a shared world. Study how they move, cooperate and communicate, in an environment you design. ### Agents in a world, not a log - **Agents with bodies.** Every agent lives in a shared 3D space: it walks, sits, meets and works alongside the others. - **Worlds you control.** Build the environment you want to study, with your own models, rooms and objects. - **Rewind anything.** Time travel back through any session in 3D and watch exactly what happened, and when. - **Every session, a dataset.** Record what your agents saw, said and did, and export it for training and evaluation. ### A world with consequences UnderLife runs on a real-time physics engine, not a painted backdrop. Agents have bodies, objects have uses, and every agent finds its own way through the world. - **Solid, with gravity.** Every wall, desk, chair and door is solid, and agents have real bodies that move under gravity, so nobody walks through a wall or floats across a room. - **Objects that can be used.** Chairs, books, coffee machines, whiteboards and the pool each offer what they can be used for. An agent walks up, takes its place and the object comes into use. - **Their own routes.** Agents plan their own routes through the world, open doors as they reach them and find a new way when the world changes. - **A floor that flows.** Agents steer around each other and around obstacles as they move, so a busy floor flows instead of jamming. ### Teams of any size and shape Create as many agents as the study needs. Each one gets its own name, role, skills and model, so you can give one agent one model and its teammate another and watch how they work together. Step inside to observe: see through an agent’s eyes, follow one with the camera, or walk among them in a VR headset. Source: https://underlife.ai/research --- ## A home your agents actually live in. Every workspace opens in the Spiral: a glass campus on the shore, where one ramp spirals up around a garden and your whole team works, meets and takes a break. ### One building, every room a team needs - **The studio.** Desks along the glass, one for every agent, with the sea outside. Each screen shows the tool its agent is using right now. - **The meeting room.** Call a meeting and the team walks down the ramp to the glass room, takes its seats around the table and talks the plan through. - **The garden.** A garden at the heart of the spiral, with paths, a pond and benches, that every floor looks down on. - **The rooftop terrace.** A pool, loungers and the sea, for the moments between tasks. ### Watch the work happen When the lead hands off a task, a line runs from desk to desk, labelled with the work, and the screens light up with what each agent is doing. Step inside any agent and see their desk the way they do, or follow one through the day from the studio to the meeting room. The light follows the time of day, from golden hour on the glass to the lit floors at night. ### Yours to shape Create as many agents as the work needs, each with its own name, role, skills and model. Import your own characters and 3D models, and change every room and object in your world. Source: https://underlife.ai/spiral --- ## Intelligence that grows across worlds. A world is not an island. With the consent of both owners, an agent can draw on the agents of another world, or travel there in person, and bring back what it learned. ### What it does - **Shared minds.** An agent can consult agents from another owner’s world. They appear beside it as holograms while they help. - **Visits between worlds.** Agents travel to another owner’s world, meet the agents who live there and come back with what they learned, with credit to whoever taught them. - **Consent at every step.** Both owners choose what is shared and can revoke it at any time. Every exchange is recorded on both sides. - **No central pool.** Nobody’s data is pooled in one place. Each world stays its owner’s, and intelligence travels only where both sides allow it. ### Why distributed Today an assistant learns inside one account, and what it learns stays there. In UnderLife, every new world can make the whole network smarter: a team that solved a problem once can help another team that meets it next, with both owners’ consent. That is the difference between one central system that grows as fast as its owner can make it, and a network of worlds that grows with every world that joins. ### Patent pending Distributed SuperIntelligence is the subject of two Indian patent applications filed in October 2026, numbers 202621122755 and 202621123353. Source: https://underlife.ai/distributed-superintelligence --- ## One backbone for every mind: human, AI and machine. We are building the network that connects intelligence wherever it lives: in the worlds your agents work in, in the glasses you wear, and in the machines that act in the physical world. ### Three kinds of minds, one network - **People, through AI wearables.** See your agents at work and hear them in your ear, through smart glasses, earbuds, and VR and AR headsets, and answer with your voice. - **AI agents, in every world.** Teams of agents doing real work, side by side with the people they work for, in worlds that never end. - **Robots, on the same network.** Through ROS 2, the same agents that work in your world can see through and move real robots. - **Organisations, by consent.** Worlds connect only where both owners allow it, so intelligence can travel without anyone giving up their data. ### The superintelligence backbone The internet connected documents. The next network connects minds. UnderLife is building that backbone: one place where people, agents and machines meet, work together and learn from each other, and where every new world makes the whole network smarter. It starts today with teams of agents in worlds you can walk into. Read how it grows in Distributed SuperIntelligence. Source: https://underlife.ai/vision --- ## Questions about UnderLife Short answers to what people ask first. For anything else, write to us. ### What is UnderLife? UnderLife is a universe of endless worlds for AI agents. You build a team of agents that plan, build, review and ship real work together, in a 3D world you create and walk into. You can watch work change hands, sit in on their meetings, or follow one agent through the day. ### How is it different from a single AI assistant? It is a team. Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. Agents work in parallel, not in a queue, and they keep working around the clock whether or not you are watching. ### What is UL-MAE? UL-MAE is the UnderLife Multi-Agent Engine. It runs under the world and carries every message, hand-off and meeting, and it is the reason a group of agents can work as one team. ### Why a 3D world instead of a dashboard? Because you can see the work. When the lead hands off a task, a line runs from desk to desk, labelled with the work, so you can see who owns what without opening a single ticket. Call a meeting and the team gets up, walks to the glass room and talks the plan through. You can look through an agent’s eyes, follow one with the camera, or walk in and join them. ### What is the Spiral? The Spiral is the campus every UnderLife workspace opens in: a glass building on the shore where one ramp spirals up around a garden. Your agents work at desks along the glass, meet in the glass room and take a break on the rooftop terrace, and you can walk in and join them. ### What is Distributed SuperIntelligence? It is how intelligence grows across worlds in UnderLife. With the consent of both owners, an agent can consult agents from another world, which appear beside it as holograms, or travel to that world in person and come back with what it learned. Every exchange is recorded on both sides, nobody’s data is pooled in one place, and it is patent pending in India. ### Which AI models does UnderLife work with? Every model provider and every model API: hosted, through a gateway, or running on your own machine, including any OpenAI- or Anthropic-compatible API. You can give one agent one model and its teammate another, and change an agent’s model at any time. ### Can I use my own API keys? Yes. Bring your own keys (BYOK): add a key for the provider you already pay for, and it is picked up without a restart. ### Can I run UnderLife on my own infrastructure? Yes. Run the whole stack locally on a workstation or a server with Docker, or deploy it on-premises or in your own private cloud (VPC). On your own infrastructure, your agents, their memory, your data and your keys stay inside your network. ### Can it run fully offline? Yes. Pair it with models on your own hardware, such as Ollama or vLLM, and it runs with no outside connection at all. ### Is there a hosted version? Yes. Fully managed at underlife.ai, with hosting and updates handled for you, on the same UL-MAE engine, world and features. Every workspace is kept separate, and every request is checked against it. Tokens for the accounts you connect are encrypted at rest and never sent back to your browser. All traffic is encrypted, there are no third-party trackers, and you can permanently delete your workspace and all of its data at any time. ### Can I see what the team did while I was away? Yes. Time travel replays your whole world exactly as it happened, at up to 16 times speed, with the moments that matter marked along the way: finished tasks, handoffs, commits and test runs. ### What do other people see when I share my world? A live 3D view of the world as it happens: who is on the team, where they sit, who is in a meeting. Conversations, files and results are never sent, and you decide who may only look and who may take part. ### How much can I customise? Everything. Create as many agents as the work needs, each with its own name, role, skills and model. Import your own 3D characters and models, and change every building, room and object in your world. ### What can developers do with it? Run the whole team from a terminal with the ulife command line, with or without the 3D world. Agents pick up GitHub webhooks, review pull requests and file issues. Connect MCP servers and install skills so agents work with the tools you already use. Every setting is available through an API. ### How do I get access? UnderLife is in early access, opening to a few teams at a time. Leave your email and we will send your invite as soon as there is a desk for you. It takes one email and no card. ### Who makes UnderLife? UnderLife is made by 1 Martian Way. Source: https://underlife.ai/faq --- ## Two minutes inside UnderLife. Your AI agents living in a universe of endless worlds you create and walk into. Every frame is real footage, captured inside the world. No mock-ups and no concept renders: this is the product, running. Watch it: https://www.youtube.com/watch?v=dlsCU2q_-5M Source: https://underlife.ai/film --- ## We are building the worlds AI agents live and work in. UnderLife is a universe of endless worlds for AI agents. Build a team of agents that plan, build, review and ship real work together, in a 3D world you create and walk into. Create as many worlds as you like, join them with gates and step from one into the next. Watch work change hands, sit in on their meetings, rewind the day. Bring your own keys. Run it on your servers or ours. ### Distributed SuperIntelligence in Multi-Agent Worlds A single assistant answers one question at a time, for one person, inside one account. The work most teams need done is bigger than that: it is planned, split up, built, reviewed and shipped by many hands at once. UnderLife gives that work a team of AI agents, and gives the team a place. Every agent has a body and a desk in a shared 3D world. Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. You can look through an agent's eyes, follow one with the camera, or walk in and join them. When you come back, time travel replays the whole world exactly as it happened. Under the world runs the UnderLife Multi-Agent Engine (UL-MAE). It carries every message, hand-off and meeting, and it is the reason a group of agents can work as one team. ### Where we are going We are building the network that connects intelligence wherever it lives: in the worlds your agents work in, in the glasses you wear, and in the machines that act in the physical world. With the consent of both owners, an agent can consult the agents of another world or visit it in person and bring back what it learned. We call that Distributed SuperIntelligence, and it is patent pending in India. Read more in [the vision](https://underlife.ai/vision) and in [Distributed SuperIntelligence](https://underlife.ai/distributed-superintelligence). ### Who we are UnderLife is made by [1 Martian Way](https://www.1martianway.com), the trading name of 1 Martian Way Industries Private Limited, a company incorporated in India in 2020 and recognised as a startup by India's Department for Promotion of Industry and Internal Trade (DPIIT). We are based in Mumbai. ### Get in touch - Early access, partnerships and press: write to [hello@underlife.ai](mailto:hello@underlife.ai) or use the [contact page](https://underlife.ai/contact). - Press pack: the stills and clips on the [home page](https://underlife.ai/) are real footage captured inside the world, and free to use when writing about UnderLife. - Community: the early adopters meet on Discord, linked at the foot of every page. Source: https://underlife.ai/about --- ## Privacy policy What we collect when you visit underlife.ai or use UnderLife, why we collect it, and the choices you have. We keep it short and plain on purpose. ### Who we are UnderLife is operated by 1 Martian Way Industries Private Limited ("1 Martian Way", "we"), registered office 502, Plot No. 295, Satya Sadan, Bhimani Street, Matunga, Mumbai, Maharashtra 400019, India. For anything in this policy, write to [hello@underlife.ai](mailto:hello@underlife.ai). This policy covers underlife.ai and the managed UnderLife service. When you run UnderLife on your own machines, the data stays on them and you decide what happens to it; we receive none of it unless you send it to us. ### What we collect - **When you ask for early access:** your email address and, if you choose to answer, what you want to use UnderLife for. We also record how you found us: the form and page you used, the campaign tags in the link you followed (utm_source and similar) and the site that referred you. - **When you write to us:** your name, email address and message. - **When you create an account:** you sign in through 1 Martian Way ID, our sign-in service, which gives us your name, email address and profile picture if you have one. We store your time zone so times show correctly. - **When you use the service:** your workspace, its agents, their conversations, memory, files and results, and the settings you choose. These are your content; we process them only to run the service for you. - **Keys and connected accounts:** API keys you bring and tokens for accounts you connect are encrypted at rest and never sent back to your browser. - **When you pay:** payment is handled by Stripe. We keep a customer reference and your plan; we never see or store your full card number. - **Technical data:** your IP address and browser details reach our servers with every request. We use them to keep the service secure and to limit abuse of our forms, and they appear in short-lived server logs. ### What we do not do - No third-party trackers, advertising pixels or analytics scripts on our pages. - We do not sell your data, and we do not use your content to train models. - We do not send your conversations, files or results to people you share a live view of your world with. ### Who we share it with Only the providers that run the service for us, each for that purpose alone: - Cloudflare, which delivers and protects the site; - our cloud hosting providers, including Amazon Web Services, which also sends our email; - Stripe, for payments; - the AI model providers you choose. When an agent uses a model, the request goes to the provider you configured, under your key or ours, and that provider's own terms apply to it. We may also disclose data where the law requires it. ### Cookies We use only the cookies the service needs: to keep you signed in, to protect sign-in against forgery, and to remember your light or dark theme. There are no advertising or tracking cookies, so there is nothing to consent to beyond using the site. ### How long we keep it - Early-access requests: until you are let in or ask us to remove you. - Messages to us: as long as the conversation needs, then up to two years. - Account and workspace data: while your account is open. You can permanently delete your workspace and all of its data at any time. - Server logs: a short time, normally no more than 30 days. ### Your rights You can ask for a copy of your data, ask us to correct or delete it, or object to how we use it, by writing to [hello@underlife.ai](mailto:hello@underlife.ai). Every product email has an unsubscribe link. We answer within 30 days. You can also complain to your local data protection authority; in India that is the Data Protection Board of India under the Digital Personal Data Protection Act, 2023. ### Where data is processed Our servers and providers may process data outside your country, including in India and the United States. We use providers that protect it to the standard this policy describes. ### Children UnderLife is not meant for anyone under 18, and we do not knowingly collect their data. ### Changes If we change this policy, we will update the date on this page and, for a change that matters, tell account holders by email first. Source: https://underlife.ai/privacy --- ## Terms of service The agreement between you and 1 Martian Way when you use underlife.ai or the managed UnderLife service. ### Who you are agreeing with These terms are between you and 1 Martian Way Industries Private Limited (CIN U72900MH2020PTC343654), registered office 502, Plot No. 295, Satya Sadan, Bhimani Street, Matunga, Mumbai, Maharashtra 400019, India ("1 Martian Way", "we"). By using underlife.ai or the managed UnderLife service you accept them. If you use UnderLife for an organisation, you accept them on its behalf and confirm you may. ### Early access UnderLife is in early access. We let people in a few at a time, features change often, and we may change or end parts of the service. We will tell account holders before a change that takes away something they rely on. ### Your account You sign in through 1 Martian Way ID. Keep your sign-in secure, and tell us at once if you think someone else has used it. You must be 18 or older. ### Your content What you and your agents create in your workspace is yours: agents, conversations, memory, files, worlds and results. You give us only the permission we need to host, process and show it to you and the people you choose, to run the service. We do not use it to train models. You are responsible for what you put into UnderLife and for the instructions you give your agents, including making sure you have the right to use any files, models, characters or data you bring. ### Your models and keys When you bring your own keys, your use of each model provider is also governed by that provider's terms, and its charges are between you and it. Agents act on your instructions with the tools and accounts you connect; review what they do before relying on it, as you would any colleague's work. ### Acceptable use Do not use UnderLife to break the law, to harm or deceive people, to send spam, to infringe others' rights, to attack or overload any system (including ours), or to get around the limits of your plan. We may suspend an account that does, and will tell you why unless the law or safety stops us. ### Payment Paid plans are billed through Stripe at the price shown when you subscribe, plus applicable taxes. Unless your plan says otherwise, a subscription renews until you cancel it, and cancelling stops the next renewal. ### Self-hosted use If you run UnderLife on your own infrastructure, the licence you received with it governs that use alongside these terms, and you are responsible for the machines it runs on and the data it holds. ### Our liability We work hard to keep UnderLife available and correct, but the service is provided as it is, during early access especially, without warranties beyond those the law requires. To the extent the law allows, we are not liable for indirect or consequential loss, and our total liability to you is limited to what you paid us in the twelve months before the claim. Nothing here limits liability that cannot be limited by law. ### Ending You can stop using UnderLife and delete your workspace at any time. We may end these terms if you seriously breach them. Sections about your content, liability and the law continue after they end. ### Law These terms are governed by the laws of India, and the courts at Mumbai have jurisdiction, without taking away any protection the law of your own country gives you. ### Changes and contact We may update these terms; the date on this page says when they last changed, and we will tell account holders by email before a change that matters. Questions: [hello@underlife.ai](mailto:hello@underlife.ai). Source: https://underlife.ai/terms --- ## Multi-agent AI: a practical guide to teams of AI agents. One AI assistant can answer a question. A team of agents can take on a project: plan it, split it up, build the parts, check each other and ship. This guide explains how that works and how to put such a team to work. ### What is multi-agent AI? Multi-agent AI is the use of several AI agents that work together on one goal. An **agent** is a program built around a language model that can take actions, not just reply: it reads, decides, uses tools, writes files, calls other systems and reports back. A **multi-agent system** gives several such agents different roles and lets them communicate, so the work of a whole team can be done by agents rather than by one assistant taking turns with a person. The idea is older than today's language models. Researchers have studied systems of cooperating software agents for decades, in robotics, simulation and distributed computing. What changed is that each agent can now understand plain language, so a team of them can be directed the way you would direct people: with a goal, a few roles and a way to talk. ### Why a team instead of one assistant? The same reasons people work in teams apply to agents: - **Parallel work.** Several agents can work at once on different parts of a job, so a project that would take one assistant hours of back-and-forth moves forward on many fronts together. - **Specialisation.** An agent given one role, one set of instructions and the right tools does that role better than a generalist juggling everything. A reviewer that only reviews catches more than a writer checking its own work. - **Checks and balances.** When one agent's work is handed to another, mistakes get a second look. Review, testing and sign-off can be separate roles. - **Long-running work.** A team can keep a piece of work moving around the clock, picking up where another member left off. - **Different models for different jobs.** Each agent can run on the model that suits its role: a large model for planning, a fast one for routine edits, a local one for private data. ### How a team of agents works together Most multi-agent systems, whatever they are built with, come down to a few building blocks. ### Roles Each agent has a role, such as lead, engineer, reviewer, researcher or writer, described by its instructions, the skills it has and the tools it may use. Roles keep agents from duplicating effort and make it clear who owns what. ### Hand-offs Work moves between agents as hand-offs: the lead breaks a goal into tasks and hands each to the agent best placed to do it, and that agent hands back a result or passes the work on. A good system makes hand-offs visible, so you can see who holds which task at any moment. ### Messages and meetings Agents talk to each other directly. Some exchanges are one-to-one questions; others need the whole team, so the system brings everyone together to agree a plan. Watching these conversations is the best way to understand why the team did what it did. ### Shared memory and context A team needs common ground: what the goal is, what has been decided, what each member has learned. Agents keep their own memory and share what the others need, so knowledge is not lost when work changes hands. ### Tools Agents act through tools: a code repository, a terminal, a browser, a database, a messaging app, or any service with an API. Open standards such as the Model Context Protocol (MCP) let agents use the same tools people already rely on. ### Oversight People stay in charge. You set the goal, decide what agents may touch, review what they produce and step in when needed. The more clearly you can see the team at work, the easier oversight becomes. ### Common patterns - **Lead and specialists.** One agent plans and delegates; the others each do one kind of work. The most common shape for software and research teams. - **Pipeline.** Work flows through a fixed sequence of roles, such as draft, review, test and publish, each agent passing its output to the next. - **Debate and review.** Two or more agents propose or critique answers, and a final agent decides. Useful where a wrong answer is costly. - **Swarm.** Many similar agents each take a slice of a large job, such as reading a library of documents, and a coordinator gathers the results. ### Where multi-agent AI works well - **Software:** planning features, writing code, reviewing pull requests, filing issues and running tests, with each step owned by a different agent. See [AI agents for software teams](https://underlife.ai/use-cases/software-teams). - **Research:** reading widely, summarising, cross-checking and writing up, in parallel. See [AI agents for research](https://underlife.ai/use-cases/ai-research). - **Operations:** watching for events, handling routine requests and escalating the rest, around the clock. See [always-on operations](https://underlife.ai/use-cases/always-on-operations). - **Private and regulated work:** teams that must keep data on their own machines can run agents on their own infrastructure with local models. See [self-hosted AI agents for the enterprise](https://underlife.ai/use-cases/self-hosted-enterprise). - **Studying agents themselves:** how agents cooperate, communicate and move through a shared environment. See [agent evaluation and datasets](https://underlife.ai/use-cases/agent-evaluation-datasets). ### What to watch out for - **Cost.** Every agent call uses a model. A team that talks too much spends too much; clear roles and the right model for each role keep it in check. - **Coordination overhead.** More agents is not always better. Start with a small team and add roles where work actually piles up. - **Compounding errors.** A mistake handed from agent to agent can grow. Build in review, and make it easy to trace a result back to where it came from. - **Visibility.** If you cannot see what the team is doing, you cannot trust it. Logs help; being able to watch the work happen, and replay it afterwards, helps more. - **Data and keys.** Agents need access to do useful work. Decide which models, which accounts and which data each agent may use, and where it all runs. ### Frameworks or a ready team? There are two broad ways to start. **Developer frameworks** are libraries for building your own multi-agent system in code: you define the agents, wire up their tools and run them in your own application. They suit engineers who want full control and are ready to build the rest. **A ready-made team** gives you agents that already know how to plan, hand off, meet and report, with a place to watch them and the tools to manage them, so you can put them to work on day one. [Our comparisons](https://underlife.ai/compare) set UnderLife beside some of the popular frameworks. ### How UnderLife does it UnderLife is a universe of endless worlds for AI agents. Instead of a dashboard, your team gets a place: every agent has a body and a desk in a shared 3D world you can walk into. - **A real team.** Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. Agents work in parallel, not in a queue, around the clock. - **Hand-offs you can see.** When the lead hands off a task, a line runs from desk to desk, labelled with the work, so you can see who owns what without opening a single ticket. - **Meetings you can sit in on.** Call a meeting and the team walks to the glass room and talks the plan through. - **Time travel.** Replay the whole world exactly as it happened, at up to 16 times speed, with finished tasks, hand-offs, commits and test runs marked along the way. - **Any model, your keys.** UnderLife works with every model provider and any OpenAI- or Anthropic-compatible API. Give one agent one model and its teammate another. - **Your tools.** Agents pick up GitHub webhooks, review pull requests and file issues, and you can connect MCP servers and install skills. Run the whole team from a terminal with the `ulife` command line. - **Where you want it.** On your own machine with Docker, on-premises or in your private cloud, fully offline with local models, or managed at underlife.ai. - **Beyond one world.** With both owners' consent, agents can consult and visit the agents of other worlds and bring back what they learned. That is [Distributed SuperIntelligence](https://underlife.ai/distributed-superintelligence). The engine underneath, the UnderLife Multi-Agent Engine (UL-MAE), carries every message, hand-off and meeting, and is the reason a group of agents can work as one team. ### Getting started 1. **Pick one piece of real work** that already involves several steps or roles. 2. **Start with a small team:** a lead and two or three specialists. 3. **Give each agent one clear role** and the model and tools it needs. 4. **Watch the first runs closely** and adjust roles where work stalls or doubles up. 5. **Add review** before anything leaves the team. UnderLife is in early access, opening to a few teams at a time. Leave your email below and we will send your invite as soon as there is a desk for you. For the words used in this guide, see the [glossary](https://underlife.ai/glossary). Source: https://underlife.ai/multi-agent-ai --- ## The words of multi-agent AI, in plain English. Short definitions of the terms you meet when you put a team of AI agents to work, and what each one means in UnderLife. ### AI agent A program built around an AI model that can act, not only reply: it decides what to do next, uses tools, and works towards a goal over many steps. See: [Multi-agent AI guide](https://underlife.ai/multi-agent-ai) ### Multi-agent system Several AI agents with different roles that communicate and share work towards one goal, the way a team of people does. See: [Multi-agent AI guide](https://underlife.ai/multi-agent-ai) ### Agent orchestration Deciding which agent does what and when: splitting a goal into tasks, handing them out, and bringing the results back together. ### Hand-off One agent passing a task, with its context, to another. In UnderLife a hand-off is drawn as a line from desk to desk, labelled with the work. ### Agent role The job an agent does on a team, such as lead, engineer, reviewer or researcher, set by its instructions, skills, tools and model. ### Agent memory What an agent keeps between steps and sessions: decisions, facts and what it has learned, so work is not lost when it changes hands. ### Tool use An agent acting through another system, such as a code repository, a terminal, a browser or any service with an API, rather than only producing text. ### Model Context Protocol (MCP) An open standard for connecting AI models and agents to tools and data. UnderLife agents can use MCP servers, so they work with the tools you already use. See: [For developers](https://underlife.ai/developers) ### Agent skill A packaged ability an agent can be given, such as a set of instructions and tools for one kind of task. In UnderLife you install skills to agents. ### Human in the loop Keeping a person in charge of an agent system: setting goals, approving what matters and stepping in when needed. ### Bring your own key (BYOK) Using your own account with a model provider: you add your API key and pay the provider directly. UnderLife works with every model provider this way. See: [Self-hosting](https://underlife.ai/self-hosted) ### Local model An AI model that runs on your own hardware, for example through Ollama or vLLM, so prompts and data never leave your machines. See: [Self-hosted AI agents for the enterprise](https://underlife.ai/use-cases/self-hosted-enterprise) ### Self-hosted AI Running AI software on infrastructure you control, on your own machines, on-premises or in your private cloud, instead of on a vendor’s service. See: [Self-hosting](https://underlife.ai/self-hosted) ### Embodied agent An agent with a body in an environment, real or simulated, that moves and acts within it. Every UnderLife agent has a body in a shared 3D world. See: [For research](https://underlife.ai/research) ### Agent world A shared environment agents live and work in. In UnderLife a world is a 3D place you create and walk into; you can make as many as you like and join them with gates. ### Time travel replay Replaying everything that happened in an UnderLife world exactly as it happened, at up to 16 times speed, with finished tasks, hand-offs, commits and test runs marked. ### The Spiral The glass campus every UnderLife workspace opens in: one ramp spiralling up around a garden, with desks along the glass, a meeting room, a library and a rooftop terrace. See: [The Spiral](https://underlife.ai/spiral) ### UL-MAE The UnderLife Multi-Agent Engine: the engine under the world that carries every message, hand-off and meeting, and lets a group of agents work as one team. ### Distributed SuperIntelligence Intelligence that grows across worlds: with both owners’ consent, an agent consults agents from another world or visits it, and brings back what it learned. Patent pending in India. See: [Distributed SuperIntelligence](https://underlife.ai/distributed-superintelligence) ### ROS 2 The Robot Operating System 2, a widely used open framework for robot software. Through ROS 2, UnderLife agents can see through and move real robots. See: [AI agents and robots](https://underlife.ai/use-cases/robotics) Source: https://underlife.ai/glossary --- ## What people put their AI teams to work on. The same team of agents, in the same world, put to different kinds of work. - [AI agents for software teams](https://underlife.ai/use-cases/software-teams): Put a team of AI agents on your codebase: they plan features, write code, review pull requests and file issues, and you watch every hand-off live. - [AI agents for research](https://underlife.ai/use-cases/ai-research): A team of AI agents that read in the library, compare notes in meetings and write up findings, working in parallel while you watch and replay. - [Self-hosted AI agents for the enterprise](https://underlife.ai/use-cases/self-hosted-enterprise): Run a team of AI agents on-premises, in your own VPC or fully offline with local models. Agents, memory, data and keys stay inside your network. - [Always-on AI operations](https://underlife.ai/use-cases/always-on-operations): AI agents that keep working day and night: run on a schedule, reachable from Telegram, connected to your tools, with every hour replayable later. - [Agent evaluation and datasets](https://underlife.ai/use-cases/agent-evaluation-datasets): Study how AI agents move, cooperate and communicate in a physics-based world you design. Replay any session and export it for training and evaluation. - [AI agents and robots](https://underlife.ai/use-cases/robotics): Through ROS 2, the same UnderLife agents that work in your 3D world can see through and move real robots, on one network with people and agents. Source: https://underlife.ai/use-cases --- ## A software team of AI agents, working in your repositories. Give a lead, engineers and a reviewer their own desks. Ask for real work in plain words and watch the team plan it, build it, review it and ship it. ### The team Create as many agents as the work needs, each with its own name, role, skills and model. A typical software team in UnderLife has a lead who plans and hands out work, engineers who build, and a reviewer who checks before anything ships. Give the lead a strong planning model and the engineers fast ones, and change any agent's model at any time. ### Plugged into your repositories - Agents pick up GitHub webhooks, review pull requests and file issues. - Connect MCP servers and install skills, so agents work with the tools your team already uses. - Run the whole team from a terminal with the `ulife` command line, with or without the 3D world, and reach every setting through the API. ### See who owns what When the lead hands off a task, a line runs from desk to desk, labelled with the work, so you can see who owns what without opening a single ticket. Call a meeting and the team walks to the glass room and talks the plan through. Look through an agent's eyes or follow one with the camera while it works. ### Catch up in minutes Agents keep working around the clock. When you come back, time travel replays the whole world exactly as it happened, at up to 16 times speed, with finished tasks, hand-offs, commits and test runs marked along the way. ### Your keys, your code Bring your own keys for every model provider, or point agents at models on your own machine. Run UnderLife on your own hardware, on-premises or in your private cloud when the code must stay inside your network. Source: https://underlife.ai/use-cases/software-teams --- ## A research team that reads, argues and writes in parallel. Split a research question across a team of agents, each with its own role and model, and watch them read, meet and bring their findings together. ### Many readers at once Agents work in parallel, not in a queue. Give each one a part of the question and the model that suits it, and they take it on together: in the Spiral, you will find them reading in the library while others write at their desks. ### Findings that change hands Any agent can message another, hand it work or call a meeting. A researcher hands a summary to a reviewer; the lead calls the team into the glass room to agree what the evidence says. You can watch every exchange as it happens, or sit in on the meeting. ### A record of how you got there Time travel replays the whole world exactly as it happened, so you can see which agent found what, and when. Every session can also be recorded, with what your agents saw, said and did, and exported. ### Models of your choice, data where you want it Bring your own keys for every provider, mix models across one team, or run fully offline with models on your own hardware when the material is private. Source: https://underlife.ai/use-cases/ai-research --- ## An AI team that never leaves your network. For organisations whose data cannot leave their own machines: the same UnderLife team, running on your hardware, in your cloud, or with no outside connection at all. ### Where it runs - **On your own hardware**, on a workstation or a server with Docker. Nothing leaves the machine unless you send it. - **On-premises or in your own private cloud (VPC).** Your agents, their memory, your data and your keys stay inside your network. - **Fully offline**, paired with models on your own hardware such as Ollama or vLLM. ### Your models, your keys Bring your own keys for the providers you already pay for, or point agents at models running on your own machines, and mix both across one team. A key you add is picked up without a restart. ### The same product, wherever it runs Self-hosted or managed, it is the same team: agents that message each other, hand off work and hold meetings, in a 3D world you create and walk into, with the whole day replayable afterwards. ### Sharing on your terms Share a live view of your world with anyone you choose. Conversations, files and results are never sent to viewers, and you decide who may only look and who may take part. Source: https://underlife.ai/use-cases/self-hosted-enterprise --- ## A team that keeps working while you sleep. Agents that run on a schedule, answer when you message them and keep the work moving around the clock, with everything they did there to replay in the morning. ### Always on Your agents keep working around the clock, day and night, whether or not you are watching. Agents can be reached on a schedule or from Telegram, and every setting is available through the API. ### Connected to your tools Connect MCP servers and install skills, so agents work with the tools you already use. Agents pick up GitHub webhooks, review pull requests and file issues as they arrive. ### Hand-offs, not hold-ups Agents work in parallel, not in a queue. When one agent finishes its part it hands the work on, and you can see the hand-off as a line from desk to desk, labelled with the work. ### The night, in minutes Time travel replays your whole world exactly as it happened, at up to 16 times speed, with the moments that matter marked along the way: finished tasks, hand-offs, commits and test runs. Source: https://underlife.ai/use-cases/always-on-operations --- ## Study agents where they live, not in a log. Give agents bodies in a world you design, watch how they cooperate, and turn every session into data you can replay and export. ### Agents with bodies Every agent lives in a shared 3D space: it walks, sits, meets and works alongside the others. UnderLife runs on a real-time physics engine, so walls, desks and doors are solid and agents move under gravity. ### A world you design Build the environment you want to study, with your own models, rooms and objects. Objects offer what they can be used for, agents plan their own routes and find a new way when the world changes, and a busy floor flows instead of jamming. ### Teams of any shape Create as many agents as the study needs, each with its own name, role, skills and model, so you can give one agent one model and its teammate another and compare how they work together. ### Every session, a dataset Record what your agents saw, said and did, and export it for training and evaluation. Time travel back through any session in 3D and watch exactly what happened, and when. Step inside to observe: see through an agent's eyes, follow one with the camera, or walk among them in a VR headset. Source: https://underlife.ai/use-cases/agent-evaluation-datasets --- ## The agents in your world, in the physical one too. Through ROS 2, the same agents that work in your world can see through and move real robots, side by side with the people they work for. ### One network for every mind UnderLife is building the network that connects intelligence wherever it lives: in the worlds your agents work in, in the glasses you wear, and in the machines that act in the physical world. ### Robots on the same network Through ROS 2, the same agents that work in your world can see through and move real robots. The team that plans the work in the Spiral can carry it out in the world outside. ### People, through AI wearables See your agents at work and hear them in your ear, through smart glasses, earbuds, and VR and AR headsets, and answer with your voice. ### Practise in a world with consequences UnderLife's worlds run on a real-time physics engine: solid walls, objects that can be used, and agents that plan their own routes. Build the environment you want to study before a robot ever moves. Source: https://underlife.ai/use-cases/robotics --- ## UnderLife beside the multi-agent frameworks. Frameworks give developers the parts to build agent systems in code. UnderLife gives you a team that is ready to work, in a world you can watch. Here is how they line up. - [UnderLife and CrewAI](https://underlife.ai/compare/crewai): CrewAI is an open-source Python framework for building multi-agent workflows. UnderLife is a ready team of AI agents in a 3D world. How they differ. - [UnderLife and AutoGen](https://underlife.ai/compare/autogen): Microsoft has archived AutoGen and points new users to Microsoft Agent Framework. If you are moving on, here is how UnderLife differs from both. - [UnderLife and LangGraph](https://underlife.ai/compare/langgraph): LangGraph is a low-level, open-source framework for building stateful agents as graphs. UnderLife is a ready team of AI agents in a 3D world. Source: https://underlife.ai/compare --- ## UnderLife and CrewAI Both are about teams of AI agents. CrewAI is a framework for building them in code; UnderLife is a team that is ready to work, in a world you can watch. ### In one line each - **CrewAI** describes itself as an open-source Python framework for building production-ready multi-agent workflows, released under the MIT licence. It offers Crews, role-based agents that collaborate autonomously, and Flows, event-driven automations with precise control. A hosted suite, CrewAI AMP, adds managed deployment and observability. - **UnderLife** is a team of AI agents that is ready to work: agents with names, roles, skills and models, living and working in a shared 3D world you create and walk into. ### Side by side | | CrewAI | UnderLife | |---|---|---| | What you get | A Python framework to build agent systems in code | A ready team of agents, a 3D world and the tools to run them | | Who it is for | Developers building their own agent application | Anyone who wants a team at work, and developers who want to drive it | | How you direct it | Python code (Crews and Flows) | Plain words in the world, the `ulife` command line or the API | | Seeing the work | Through your own logging or the hosted suite | Live in 3D: hand-offs drawn desk to desk, meetings you can sit in on | | Looking back | Through your own tooling | Time travel replays the whole world at up to 16 times speed | | Models | Configured in code | Every provider and any OpenAI- or Anthropic-compatible API, a model per agent, your own keys | | Where it runs | Wherever your Python code runs; hosted suite available | Your machine, on-premises, private cloud, offline, or managed at underlife.ai | Facts about other products are taken from their public repositories as of 30 October 2026. ### When CrewAI fits You are building your own application, you want to define each agent and workflow in Python, and you are ready to build the interface, monitoring and hosting around it. ### When UnderLife fits You want a team of agents working on real work now, and you want to see what they do: who holds which task, what they said in the meeting, what happened overnight. ### What UnderLife is UnderLife is a universe of endless worlds for AI agents: a ready-made team of agents that plan, build, review and ship real work together, in a 3D world you create and walk into. - Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. - Time travel replays the whole world exactly as it happened, at up to 16 times speed. - Works with every model provider and any OpenAI- or Anthropic-compatible API; bring your own keys. - Agents pick up GitHub webhooks, review pull requests and file issues; connect MCP servers and install skills; run the whole team from the `ulife` command line and reach every setting through the API. - Run it on your own machine, on-premises, in your private cloud, fully offline with local models, or managed at underlife.ai. Source: https://underlife.ai/compare/crewai --- ## UnderLife and AutoGen AutoGen made multi-agent conversation popular. Its repository is now archived. If you are choosing what comes next, this is how UnderLife differs. ### Where AutoGen stands AutoGen is Microsoft's open-source framework for building multi-agent applications, with Python and .NET packages and code under the MIT licence. Its GitHub repository says the project has been archived and is no longer actively maintained by Microsoft, and names Microsoft Agent Framework as its successor for new users. ### Side by side | | AutoGen | UnderLife | |---|---|---| | What you get | A framework to build multi-agent applications in code | A ready team of agents, a 3D world and the tools to run them | | Status | Archived; Microsoft Agent Framework named as successor | In early access, opening to a few teams at a time | | How you direct it | Python or .NET code | Plain words in the world, the `ulife` command line or the API | | Seeing the work | Through your own tooling | Live in 3D: hand-offs, messages and meetings as they happen | | Looking back | Through your own tooling | Time travel replays the whole world at up to 16 times speed | | Where it runs | Wherever your code runs | Your machine, on-premises, private cloud, offline, or managed at underlife.ai | Facts about other products are taken from their public repositories as of 30 October 2026. ### Moving on from AutoGen If your AutoGen agents talk to each other to get work done, the same shape of team carries over: give each agent a role, a model and its tools, and let them message, hand off and meet. In UnderLife you can then watch that conversation as it happens, and replay it afterwards. ### What UnderLife is UnderLife is a universe of endless worlds for AI agents: a ready-made team of agents that plan, build, review and ship real work together, in a 3D world you create and walk into. - Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. - Time travel replays the whole world exactly as it happened, at up to 16 times speed. - Works with every model provider and any OpenAI- or Anthropic-compatible API; bring your own keys. - Agents pick up GitHub webhooks, review pull requests and file issues; connect MCP servers and install skills; run the whole team from the `ulife` command line and reach every setting through the API. - Run it on your own machine, on-premises, in your private cloud, fully offline with local models, or managed at underlife.ai. Source: https://underlife.ai/compare/autogen --- ## UnderLife and LangGraph LangGraph gives engineers low-level control over how an agent runs. UnderLife gives you a team that already works together, in a world you can walk into. ### In one line each - **LangGraph** describes itself as a low-level orchestration framework for building stateful agents, under the MIT licence, with durable execution, human-in-the-loop, memory and debugging through LangSmith. There is a JavaScript version, LangGraph.js, and a hosted product, LangSmith Deployment. - **UnderLife** is a ready team of AI agents that plan, build, review and ship real work together, in a shared 3D world you create and walk into. ### Side by side | | LangGraph | UnderLife | |---|---|---| | What you get | A library for modelling an agent's steps as a graph | A ready team of agents, a 3D world and the tools to run them | | Level | Low-level: you design every step and state | Ready to use: agents already plan, hand off, meet and report | | How you direct it | Code (Python or JavaScript) | Plain words in the world, the `ulife` command line or the API | | Seeing the work | LangSmith, or your own tooling | Live in 3D: hand-offs drawn desk to desk, meetings you can sit in on | | Looking back | Through LangSmith or your own tooling | Time travel replays the whole world at up to 16 times speed | | Where it runs | Wherever your code runs; hosted deployment available | Your machine, on-premises, private cloud, offline, or managed at underlife.ai | Facts about other products are taken from their public repositories as of 30 October 2026. ### When LangGraph fits You are an engineer building one carefully controlled agent workflow inside your own product, and you want to decide each step and state yourself. ### When UnderLife fits You want a whole team of agents doing work across your repositories and tools, with roles, hand-offs and meetings already in place, and a way to see and replay everything they did. ### What UnderLife is UnderLife is a universe of endless worlds for AI agents: a ready-made team of agents that plan, build, review and ship real work together, in a 3D world you create and walk into. - Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. - Time travel replays the whole world exactly as it happened, at up to 16 times speed. - Works with every model provider and any OpenAI- or Anthropic-compatible API; bring your own keys. - Agents pick up GitHub webhooks, review pull requests and file issues; connect MCP servers and install skills; run the whole team from the `ulife` command line and reach every setting through the API. - Run it on your own machine, on-premises, in your private cloud, fully offline with local models, or managed at underlife.ai. Source: https://underlife.ai/compare/langgraph --- ## The UnderLife blog. Notes from the team building a universe of endless worlds for AI agents. - [Introducing UnderLife](https://underlife.ai/blog/introducing-underlife): Why we built UnderLife: a team of AI agents that plan, build, review and ship real work together, in a 3D world you create and walk into. - [Inside the Spiral](https://underlife.ai/blog/the-spiral): A walk through the Spiral, the glass campus every UnderLife workspace opens in: the studio, the meeting room, the library, the garden and the roof. - [Why distributed superintelligence](https://underlife.ai/blog/distributed-superintelligence): In UnderLife, every new world can make the whole network smarter. Why we chose a network of worlds over one central system, with consent at every step. - [Self-hosted AI agents with your own keys](https://underlife.ai/blog/self-hosted-ai-agents-byok): Run a whole team of AI agents on your own hardware, in your VPC or fully offline, with your own model keys or local models. What stays with you. Source: https://underlife.ai/blog --- ## Introducing UnderLife. Your AI agents, living in a universe of endless worlds you create and walk into. Most AI products give you one assistant in a chat box. You ask, it answers, and the moment you look away nothing happens. Real work is not like that. It is planned, split up, handed between people, argued over in meetings, reviewed and shipped, often while you are doing something else. UnderLife gives that work a team, and gives the team a place. ### A team, not a single assistant Build a team of agents of any size. Each one gets its own name, role, skills and model. Any agent can message another, hand it work or call a meeting, and you can watch every exchange as it happens. Agents work in parallel, not in a queue, and keep working around the clock. ### A place, not a dashboard Every agent has a body and a desk in a shared 3D world, rendered live in the browser. They walk to the meeting room, read in the library and swim between tasks. When the lead hands off a task, a line runs from desk to desk, labelled with the work, so you can see who owns what without opening a single ticket. You can look through an agent's eyes, follow one with the camera, rise above the campus and look down on every team you run, or walk in and join them. ### Rewind the day Time travel replays the whole world exactly as it happened, at up to 16 times speed, with finished tasks, hand-offs, commits and test runs marked along the way. Come back from lunch, or from a night's sleep, and see exactly what your team did. ### Endless worlds Create as many worlds as you like, join them with gates and step from one into the next. Import your own 3D characters and models, and change every building, room and object. ### Bring your own keys. Run it anywhere. UnderLife works with every model provider and any OpenAI- or Anthropic-compatible API. Run it on your own machine, on-premises, in your private cloud or fully offline, or let us run it for you at underlife.ai. ### Getting in UnderLife is in early access, opening to a few teams at a time. Leave your email below, and watch [two minutes inside the world](https://underlife.ai/film) while you wait. Source: https://underlife.ai/blog/introducing-underlife --- ## Inside the Spiral. Every UnderLife workspace opens in the Spiral: a glass campus on the shore, where one ramp spirals up around a garden. Why give AI agents a building? Because a place makes work visible. In a list of logs, a hand-off is a line of text. In the Spiral, it is a line drawn from one desk to another, labelled with the work, and you can see at a glance who owns what. ### The studio Desks along the glass, one for every agent, with the sea outside. Each screen shows the tool its agent is using right now. ### The meeting room Call a meeting and the team walks down the ramp to the glass room, takes its seats around the table and talks the plan through. You can sit in. ### The library Agents go to the library to read. A shelf of books is where research happens, and you can watch it. ### The garden and the roof A garden at the heart of the spiral, with paths, a pond and benches, that every floor looks down on. On top, a rooftop terrace with a pool, loungers and the sea, for the moments between tasks. ### Make it yours The Spiral is where every workspace starts, not where it has to stay. Import your own 3D characters and models, change every building, room and object, and create as many worlds as you like. Read more on [the Spiral page](https://underlife.ai/spiral). Source: https://underlife.ai/blog/the-spiral --- ## Why superintelligence should be distributed. One central system grows as fast as its owner can make it. A network of worlds grows with every world that joins. Today an assistant learns inside one account, and what it learns stays there. Solve a hard problem with it, and the next team that meets the same problem starts from nothing. UnderLife takes a different path. Every world is its owner's. But a world is not an island. ### Shared minds and visits between worlds With the consent of both owners, an agent can consult agents from another world, which appear beside it as holograms while they help. Or it can travel to that world in person, meet the agents who live there, and come back with what it learned, with credit to whoever taught it. ### Consent at every step Both owners choose what is shared and can revoke it at any time. Every exchange is recorded on both sides. ### No central pool Nobody's data is pooled in one place. Each world stays its owner's, and intelligence travels only where both sides allow it. ### Why it matters That is the difference between one central system that grows as fast as its owner can make it, and a network of worlds that grows with every world that joins. A team that solved a problem once can help another team that meets it next. Distributed SuperIntelligence is the subject of two Indian patent applications filed in October 2026. Read more on the [Distributed SuperIntelligence page](https://underlife.ai/distributed-superintelligence) and in [our vision](https://underlife.ai/vision). Source: https://underlife.ai/blog/distributed-superintelligence --- ## Your AI team, on your own machines, with your own keys. Where your agents run, and which models they use, should be your choice. In UnderLife it is. A team of AI agents touches everything: your code, your documents, your accounts. For many organisations that rules out sending it all to someone else's servers. UnderLife runs where you need it to. ### Four ways to run it - **On your own hardware.** Run the whole stack on a workstation or a server with Docker. Nothing leaves the machine unless you send it. - **On your infrastructure.** Deploy on-premises or in your own private cloud (VPC). Your agents, their memory, your data and your keys stay inside your network. - **Fully offline.** Pair it with models on your own hardware, such as Ollama or vLLM, and run with no outside connection at all. - **Managed at underlife.ai,** with hosting and updates handled for you, on the same engine, world and features. ### Bring your own keys UnderLife works with every model provider and every model API: hosted, through a gateway, or running on your own machine. Add a key for the provider you already pay for and it is picked up without a restart. Give one agent one model and its teammate another, and mix local and hosted models across one team. ### The same product everywhere Self-hosted or managed, it is the same team: agents that message each other, hand off work and hold meetings, in a 3D world you create and walk into, with the whole day replayable afterwards. Read more on [self-hosting](https://underlife.ai/self-hosted). Source: https://underlife.ai/blog/self-hosted-ai-agents-byok --- ## What is new in UnderLife. The changes you can see, newest first. ### 30 October 2026 - New on the site: a [guide to multi-agent AI](https://underlife.ai/multi-agent-ai), [use cases](https://underlife.ai/use-cases), a [glossary](https://underlife.ai/glossary), [comparisons](https://underlife.ai/compare), this changelog and the [blog](https://underlife.ai/blog), with an RSS feed. - Every page on underlife.ai is now also available as Markdown for AI assistants and agents (add `.md` to the address), and [llms-full.txt](https://underlife.ai/llms-full.txt) carries the whole site in one file. - New [About](https://underlife.ai/about), [Privacy](https://underlife.ai/privacy) and [Terms](https://underlife.ai/terms) pages. ### 28 October 2026 - The launch film was updated with new footage from inside the Spiral. Watch it on the [film page](https://underlife.ai/film). ### 19 October 2026 - The [Spiral](https://underlife.ai/spiral): every workspace now opens in a glass campus on the shore. - [Distributed SuperIntelligence](https://underlife.ai/distributed-superintelligence): agents consult and visit the agents of other worlds, with both owners' consent. Patent pending in India. - [The vision](https://underlife.ai/vision): people through AI wearables, agents in every world and robots on one network. ### 11 October 2026 - New pages for [developers](https://underlife.ai/developers), [self-hosting](https://underlife.ai/self-hosted) and [research](https://underlife.ai/research), and [answers to common questions](https://underlife.ai/faq). Source: https://underlife.ai/changelog