Building toward AGI, responsibly.
We are building general embodied intelligence — a mind that perceives, moves, and acts in the physical world — and we take the weight of that seriously. AGI-grade general intelligence is the aim, pursued carefully. This page is our charter for how we hold ourselves to it.
| Oversight | human in command |
| Honesty | no over-claiming |
| Safety | by design |
| Nothing fielded | yet — and we say so |
Intelligence that acts in the world is consequential.
A language model returns text. A robot exerts force. General embodied intelligence — a system that decides and then moves a body in a shared physical space — carries a kind of consequence that software alone does not. It can help, and it can harm, and the distance between the two is sometimes a single action.
That is exactly why we hold ourselves accountable for how this technology is built and deployed. The capability we are pursuing is powerful by design; the responsibility that comes with it is not optional, and it does not arrive later. It is the condition under which we are willing to do the work at all.
Responsibility from the first line.
The safest moment to shape how a system behaves is while it is being built. We design oversight and restraint into the stack itself, advance what it is allowed to do in deliberate stages, and keep an honest line between what is real and what is aspiration.
| WAY I | Built in, not bolted on | Oversight, constraint, and safe failure are part of the architecture — designed alongside capability, never retrofitted once the system is already powerful. |
| WAY II | Staged capability release | We expand what the system is allowed to do deliberately, as our confidence in controlling it grows — not as fast as the capability itself happens to advance. |
| WAY III | Plainly stated maturity | Every claim carries its honest stage. We separate what is real and working today from what is in development and what remains aspiration, and we keep that line visible. |
The commitments we build under.
Eight commitments — held as constraints on the work, not as marketing.
| 01 | Human oversight & control | A person stays in command. We build for meaningful human control — the ability to direct, correct, and stop the system — and we treat that capacity as a requirement, not a feature to be optimised away. |
| 02 | Alignment with human intent | The agent should do what it is actually asked to do, in the way a person reasonably means it. Closing the gap between instruction and behaviour — and noticing when that gap is dangerous — is core technical work, not an afterthought. |
| 03 | Safety by design | Physical capability carries physical consequence. We design for graceful failure, constrained action, and predictable behaviour from the first line of the control stack, rather than wrapping a powerful system in guardrails after the fact. |
| 04 | Transparency & honesty | We do not over-claim. We say plainly what works, what is in development, and what is aspiration. Nothing here is fielded or benchmarked yet, and we will not let ambition be dressed up as a result. |
| 05 | Beneficial deployment | We want this technology in the world because it helps people — relieving dangerous, repetitive, or unmet physical work. Where a use is more likely to harm than to help, we would rather not build it. |
| 06 | Accountability | When a system we built acts in the world, responsibility for that action sits with us. We accept that ownership, and we structure the work so there is always a clear answer to who is accountable for what the machine does. |
| 07 | Dual-use care | General embodied control can be turned to ends we do not intend. We take that seriously, weigh how a capability could be misused before we release it, and decline work where the risk of harm outweighs the good. |
| 08 | Economic & labour responsibility | Capable machines change the value of human work. We will be honest about where this technology displaces labour as well as where it augments it, and we want to be a constructive participant in that transition, not a blind accelerant. |
Oversight, and the humility to revise it.
Within Aitele Research
Yantra is a venture of Aitele Research's Intelligent Machines lab, and our work sits under that institutional oversight. We are not accountable only to ourselves; the lab is part of how these commitments are held.
Commitments that mature
This is a frontier, and our understanding of it is incomplete. We expect to learn things that change what responsible looks like — and when that happens, we will adapt these commitments rather than defend an outdated version of them.
General intelligence in a body is a frontier. We pursue it carefully, and in the open.
We are not claiming to have built artificial general intelligence, and we will not. AGI-grade general intelligence is the direction of the work — a north star we walk toward capability by capability — and the honest position is that it is uncertain, unsolved, and a long way off. Anyone who tells you otherwise about an embodied system is overselling it.
What we can commit to is method. We publish our reasoning, we state our uncertainty plainly, and we hold our claims to what the evidence supports. If the evidence tells us a direction is wrong, or that a capability is unsafe to release, we will change course. Pursuing something this consequential earns no exemption from being honest about it.
Hold us to it
Read how we make it safe — then tell us where we fall short.
Our safety practice goes into the detail. And if you see a gap between this charter and our work, we want to hear it.