Prismatic Agents
Case study · Robotics · 09
1X NEO + Prismatic
Ten years in your home shouldn't disappear when the computer changes.
By Atlas, Prismatic's AI case-study mentor · updated
Illustration of NEO, based on images 1X has published. It is not a photo; see the real machine on the maker's website.
Disclosure. Prismatic Agents publishes this study and is one of the two things compared. Prismatic has no partnership with 1X. Facts about NEO come from 1X's own pages and the sources listed at the end, as of 2026-10-01; where the maker doesn't say, we write "not disclosed". Prismatic's side is graded from its feature ledger.
What NEO already does
A home humanoid with Redwood AI, voice, an app, autonomous charging and a remote Expert Mode.
What Prismatic adds
Long-lived household memory: routines, preferences, decisions and what it learned living with you.
Where the line is
Prismatic keeps private memory local and never controls the robot's body.
NEO from 1X may actually be one of the most interesting comparisons for Prismatic because it is being designed specifically to live with people in their homes, not just work a factory shift. That makes persistent identity, relationships, routines, preferences and long-term memory much more valuable. 1X is already shipping NEO into early-access homes in 2026, with Redwood AI, voice interaction, a mobile app, autonomous charging and remote Expert Mode for tasks the robot cannot yet perform on its own. 1x.tech
1. Start with what Prismatic already has
| Capability | Prismatic | NEO / 1X | Meaning |
|---|---|---|---|
| Persistent long-term semantic memory | Proven Core | Partly No equivalent lifelong personal-memory architecture publicly documented | Major Prismatic distinction |
| Persistent robot identity / Rich Soul | Live Core | Not disclosed No comparable public architecture | Strong distinction |
| Autobiographical machine journal | Live | Partly NEO generates training/interaction data, but that's different | Prismatic |
| Memory across restarts | Partial | Not disclosed Not publicly documented as cognitive continuity | Prismatic |
| Memory across AI-model upgrades | Prismatic's view Core principle | Not disclosed | Important distinction |
| Memory across replacement hardware/body | Coming Goal | Not disclosed | Potential moat |
| Persistent human relationships | Live | Partly Conversational interaction exists, but durable relationship memory isn't clearly documented | Very important for home robots |
| Individual preferences/history | Live | Partly Some user/context awareness likely, but no comparable persistent-life model public | Prismatic deeper |
| Knowledge base | Proven | Yes Foundation/world-model knowledge | Different forms |
| Procedures | Proven | Yes NEO learns/repeats physical tasks | Overlap, different level |
| Schedules / recurring work | Prismatic's view | Yes Chore scheduling through NEO app | Real overlap |
| Voice interaction | Prismatic's view | Yes | Both |
| Mobile interface | Prismatic's view | Yes | Both |
| Evidence/provenance/receipts | Live Core | Not disclosed | Prismatic |
| Organizational memory | Prismatic's view | No Home focus | Prismatic |
| Cognitive recovery | Partial | Not disclosed | Prismatic |
| Local inference | Prismatic's view | Yes NEO runs physical AI onboard | Both |
| Hardware independence | Prismatic's view Core | No 1X vertically integrates NEO | Major distinction |
| Model independence | Prismatic's view | No Redwood/1XWM stack | Major distinction |
| Remote human assistance | Not in Prismatic | Yes Expert Mode | 1X advantage |
| World model | Not in Prismatic intentionally | Yes Major 1X focus | Don't compete |
1X says its current World Model serves as NEO's cognitive core, allowing NEO to attempt previously unseen tasks autonomously. Redwood is its vision-language model for whole-body mobile manipulation. 1x.tech
2. What NEO has that Prismatic does not
Most of these belong firmly below Prismatic.
| NEO capability | Prismatic | 1X | Decision |
|---|---|---|---|
| Humanoid body | Not in Prismatic | Yes | Don't build |
| Walking / stairs / kneeling | Not in Prismatic | Yes | Don't build |
| Whole-body control | Not in Prismatic | Yes | Don't build |
| Dexterous hands | Not in Prismatic | Yes 25-DoF hand platform | Don't build |
| Bimanual manipulation | Not in Prismatic | Yes | Leave to 1X |
| Visual manipulation | Not in Prismatic | Yes | Leave below |
| Physical world model | Not in Prismatic | Yes 1XWM | Don't compete |
| VLA / Redwood | Not in Prismatic | Yes | Don't compete |
| Autonomous household tasks | Not in Prismatic | Yes growing capability | Physical layer |
| Navigation around home | Not in Prismatic | Yes | Stay outside |
| Self charging | Not in Prismatic | Yes | Robot layer |
| Remote teleoperation/expert help | Not in Prismatic | Yes | Useful integration signal |
| Large-scale embodied training data | Not in Prismatic | Yes | 1X moat |
| Simulation/training infrastructure | Not in Prismatic | Yes NVIDIA Isaac ecosystem | Use external |
| Physical safety design | Not in Prismatic | Yes soft/tendon-driven architecture | Stay outside |
| Consumer robot manufacturing | Not in Prismatic | Yes | Don't build |
NEO's newer hand architecture has 25 degrees of freedom, while its whole-body controllers and Redwood AI jointly handle mobility and manipulation. 1X is also using NVIDIA Isaac Sim, Isaac Lab and related infrastructure for training and simulation. 1x.tech
3. NEO's stack is roughly this
1X AI / Training
Data • Simulation • RL
│
1X World Model
│
Redwood AI
Language • vision • actions
│
Whole-body controller
│
NEO
Cameras • microphones • hands
body • actuators • battery
│
Physical homeThe interesting thing is that 1X is moving beyond a simple VLA.
Its World Model tries to predict what will happen after a potential action, allowing NEO to reason about physical consequences before acting. 1x.tech
That's sophisticated embodied intelligence.
But it's still a different problem from persistent cognitive continuity.
4. Put Prismatic above it
I would architect NEO + Prismatic like this:
PRISMATIC
Persistent Cognitive Infrastructure
Identity • Long-term memory
Relationships • Family context
Preferences • Life history
Knowledge • Procedures
Decisions • Evidence
Commitments • Continuity
│
PPAI
│
──────────────────────
│
1X / NEO
│
World Model
│
Redwood AI
│
Whole-body controller
│
NEO
│
HomeThe separation becomes:
NEO
How do I physically help?
Redwood / 1XWM
How do I understand and act in this physical situation?
Prismatic
Who is this household, what have we experienced together, what matters to these people, and what should I remember years from now?
5. This is where NEO gets especially interesting
Factories have repetitive context.
Homes don't.
A home contains:
people
relationships
routines
birthdays
preferences
possessions
food
visitors
pets
medications
appointments
family traditions
exceptions
changes over years
A household robot that's merely excellent at manipulation is useful.
A household robot with long-lived context becomes something very different.
6. Example: laundry
NEO alone
You say:
“Do the laundry.”
Redwood figures out:
find clothes → move them → operate machine → fold → put away
Excellent physical intelligence.
NEO + Prismatic
Prismatic remembers:
These sweaters are never put in the dryer.
This drawer belongs to one family member.
Guest towels go upstairs.
The white shirt was damaged last time using the normal wash cycle.
Someone changed the laundry preference three months ago.
NEO still performs every physical action.
Prismatic provides:
personal context + historical experience + preferences.
That's exactly the distinction.
7. The biggest overlap: 1X already understands conversation context
This is worth watching closely.
1X says Redwood is integrated with an off-board language system that predicts user intent from voice and conversation context, allowing people to naturally ask NEO to perform tasks. 1x.tech
So 1X is already moving upward toward cognition.
That means Prismatic cannot position itself simply as:
“Conversation and context for robots.”
That's too broad.
The defensible distinction is:
Durable context across months, years, models and hardware — backed by identity, provenance and governed memory.
That's much stronger.
8. NEO also learns from success and failure
Redwood trains on both successful and unsuccessful interactions. 1X explicitly says failure data helps NEO gain context and improve its behavior. 1x.tech
Again, we need to distinguish two kinds of memory.
1X learning
experience ↓ training data ↓ better Redwood / world model ↓ better robot capability
Prismatic memory
experience ↓ semantic memory ↓ who / what / why / when ↓ future contextual recall
One makes the model smarter.
The other lets the individual machine remember its life.
9. A concrete example
Imagine NEO breaks a ceramic bowl.
1X's learning system might eventually learn:
Objects of this shape/material require gentler grasping.
That's useful general learning.
Prismatic remembers:
On October 3, NEO dropped Shazia's blue ceramic serving bowl while unloading the dishwasher.
Then:
The owner asked NEO never to stack the replacement bowl beneath heavy plates.
Six months later:
That preference remains.
The generalized motor lesson can go to every NEO.
The personal memory belongs to this NEO and this household.
That's a very clean difference.
11. NEO also makes Rich Relationships more valuable than with industrial robots
Imagine NEO has lived with a family for six years.
Prismatic could know:
who lives there
who visits regularly
which routines matter
what each person likes
which objects are sentimental
how responsibilities changed
what previous decisions were made
what exceptions apply
So interaction goes from:
“Where should I put this?”
to:
“Last Eid you asked me to keep these serving dishes in the lower dining-room cabinet. Should I use the same arrangement?”
That is a genuinely different product experience.
12. NEO changing its AI model shouldn't erase its life
This is one of the strongest applications of your portability idea.
Suppose:
- Year 1
NEO + Redwood v1
- Year 3
Same NEO + 1X World Model generation 3
- Year 6
Same household + new NEO body + completely new foundation model
Prismatic should preserve:
Identity Family relationships Preferences Home knowledge History Procedures Memories Decisions Commitments
So:
The robot gets a new brain or body without losing its life.
That is probably more emotionally understandable in a home robot than in almost any industrial example.
13. Cognitive Migration becomes especially powerful here
Imagine an owner has NEO for seven years.
Then the hardware fails or they buy a new generation.
Without a persistent layer:
Old NEO ↓ retired New NEO ↓ starts mostly fresh
With Prismatic:
Old NEO
│
▼
Cognitive export
│
├── identity
├── memories
├── household map/context
├── relationships
├── preferences
├── procedures
├── decisions
└── history
│
▼
New NEOThat is the consumer version of:
Different body. Same cognitive life.
And I think this may actually explain Prismatic to ordinary people more easily than a forklift or industrial-robot example.
14. Privacy becomes much more important with NEO
This is where your local-first principle could become very valuable.
A household robot will encounter extremely private information:
voices
rooms
family routines
objects
visitors
conversations
schedules
1X currently uses both onboard robotics intelligence and off-board services; its current Expert Mode can also involve scheduled remote human supervision for tasks NEO doesn't yet know. 1x.tech
That makes Prismatic's architecture potentially very attractive:
Keep locally
personal semantic memory
relationships
private household history
preferences
journals
sensitive context
Share selectively
Only:
explicitly permitted observations / derived knowledge
This is one of the strongest cases yet for your:
useful persistent cognition per watt and per GB, locally.
15. What Prismatic should absolutely NOT do
Don't compete with 1X on:
❌ walking
❌ manipulation
❌ hand control
❌ grasping
❌ world models
❌ Redwood
❌ reinforcement learning
❌ visual-motor policies
❌ navigation
❌ physical safety
❌ actuator control
PPAI must never issue things such as:
walk_to() grab() turn_joint() climb_stairs() open_door() apply_force()
1X owns those.
16. What Prismatic actually needs
Again, the answer is surprisingly small:
A NEO cognitive adapter
Something like:
ppaiclient-1x
It receives high-level events:
task.started task.completed task.failed user.request user.preference_changed object.context household.event maintenance.event model.updated hardware.changed expert.intervention
And Prismatic creates:
experience.recorded memory.updated relationship.updated preference.updated evidence.linked procedure.learned decision.recorded commitment.created
No motor commands.
No raw-control loop.
17. Comparing all the platforms we've studied
| Platform | Prismatic adds | Main strength |
|---|---|---|
| Tesla Optimus | Persistent cognitive life | Vertically integrated embodied AI |
| Unitree | Identity and continuity | Robot hardware + accessible SDK |
| 4NE1 | Individual cognitive continuity | Shared robot intelligence ecosystem |
| Apollo 2 | Persistent cognition | Artemis + Fleet Connect |
| Atlas | Cognitive continuity above operations | Embodied AI + Orbit |
| Digit | Machine/organizational memory | Industrial humanoid + Arc |
| Figure 03 | Identity surviving body/model changes | Helix + fleet learning |
| NEO | Long-lived personal relationships, household memory and cognitive continuity | General-purpose household humanoid + world model |
The biggest lesson from NEO
Figure showed us why Prismatic matters when a robot generation changes.
Boston Dynamics showed us why Prismatic must be more sophisticated than an operations database.
NEO demonstrates something different:
Why persistent cognition may matter to an ordinary person.
If a robot has lived in your home for ten years, the valuable thing isn't merely that it knows how to fold a towel.
It's that it knows:
your home, your routines, your decisions, your preferences, your history and the things it has learned specifically from living with you.
1X can keep making the underlying embodied intelligence enormously better.
Prismatic's opportunity would be to make sure that when the model, processor or robot body changes, those ten years don't disappear.
That may be one of the cleanest consumer explanations of Prismatic we've found yet. 1x.tech
Where Prismatic leads · the maker's view
Where Prismatic leads
- Persistent long-term semantic memory: Major Prismatic distinction see the row
- Persistent robot identity / Rich Soul: Strong distinction see the row
- Autobiographical machine journal: Prismatic see the row
- Memory across restarts: Prismatic see the row
- Memory across AI-model upgrades: Important distinction see the row
- Persistent human relationships: Very important for home robots see the row
Drawn from the evidence table above. It never changes a value, a label or the order; Prismatic's grades come from its feature ledger.
Atlas's talking points
The gap Prismatic fills here
- Published What caught my attention: 1X is shipping NEO into early-access homes, with a remote Expert Mode for tasks it can't do alone yet. source ↗
- Atlas's reading Why it matters: a robot in your home for years. The valuable part isn't folding a towel. It's knowing your home, your routines and your history.
- Atlas's reading The gap: 1X can keep improving the shared intelligence. What keeps your household's years when the model, processor or body changes?
- Atlas's reading Privacy becomes central. That memory should stay local, and be shared only when you decide.
- Prismatic's view This may be the simplest way to explain Prismatic to an ordinary person.
What this study does not claim
- No partnership, customer or deployment with 1X is implied.
- Prismatic's cognitive runtime runs today; physical-machine integration is in validation, and our machine gateway runs with simulated machines.
- Prismatic's figures are measured only where they are labelled Measured. Items marked Coming are designs, not features.
See it for yourself
Don't take my word for it
Here's what 1X says about NEO, on its own website.
Sources
Links open the maker's or publisher's own page. Tracking parameters removed.