Sanctuary AI Phoenix + Prismatic

1. Start with what Prismatic already has

Sanctuary now explicitly says its Physical AI is multi-embodiment, designed to work across different hardware configurations, mobility systems and end effectors—not just Phoenix. sanctuary.ai

That makes Sanctuary strategically closer to Prismatic than Tesla or Figure.

3. Sanctuary's architecture

Conceptually:

             Sanctuary AI
                 Carbon
                   │
    ┌──────────────┼──────────────┐
    │              │              │
 Reasoning      Task plans    Behaviour models
    │              │              │
 Language       Perception       Policies
    │              │              │
    └──────────────┬──────────────┘
                   │
             Robot control
                   │
      ┌────────────┼────────────┐
      │            │            │
   Phoenix    Industrial Arm    AMR
      │            │            │
      └──── Physical World ─────┘

Carbon already sits higher than many VLA systems.

Sanctuary describes it as combining modern LLMs, symbolic/logical reasoning, domain integrations, agency and auditable task/motion plans. sanctuary.ai

So simply saying:

“Prismatic supplies intelligence above the robot”

would not sufficiently distinguish Prismatic from Sanctuary.

4. Where Prismatic fits

The cleaner architecture is:

                    PRISMATIC
          Persistent Cognitive Infrastructure

 Identity • Long-term memory • Experience
 Relationships • Organizational knowledge
 Evidence • Decisions • Procedures
 Work history • Governance • Continuity
                     │
                    PPAI
                     │
                  Carbon
                     │
       Reasoning / Physical task policy
                     │
             Phoenix / other robot
                     │
              Physical world

5. Carbon already has reasoning — so Prismatic can't claim that as its moat

This comparison forces another useful refinement.

Carbon publicly includes:

• natural-language understanding
• symbolic and logical reasoning
• LLM knowledge
• goal seeking
• explainable reasoning
• auditable task plans
• fleet management

sanctuary.ai

So Prismatic's moat cannot simply be:

“reasoning for physical robots.”

Sanctuary is already doing that.

Instead:

Persistent cognitive continuity around replaceable reasoning systems.

Carbon can become much smarter.

Carbon v2 can replace Carbon v1.

A new behaviour model can replace an old one.

Phoenix can be replaced by another embodiment.

But the Prismatic identity/memory layer could remain.

6. Example: Phoenix + Prismatic

Imagine Phoenix-12 working in a manufacturing facility.

Carbon handles

Human:

“Sort these assemblies and place failed parts in inspection.”

Carbon:

• understands request,
• perceives objects,
• generates task/motion plan,
• manipulates parts,
• reacts to failures,
• completes work.

All Sanctuary.

Prismatic receives

Phoenix-12 encountered another failure on Part Type Z.

It recalls:

Four similar failures occurred over the previous six weeks.

Then:

Phoenix-04 encountered the same issue at another station.

Prismatic links:

supplier batch
quality reports
maintenance history
operator comments
previous engineering decision

It provides:

“Seven incidents correlate with supplier batch 447. Previous corrective action temporarily reduced failures.”

Human approves an investigation.

Prismatic stores:

experience → evidence → human decision → outcome → organizational memory

Carbon continues doing the physical task.

7. This reveals the difference between Behaviour Learning and Experience Memory

Sanctuary is heavily interested in behavioural data.

Its work on Large Behaviour Models is intended to let robots learn from real-world experience, using high-quality behavioural data to improve physical capabilities. sanctuary.ai

So:

Sanctuary

Robot experience
      ↓
Behaviour data
      ↓
Training
      ↓
Better Carbon / policy
      ↓
Better physical performance

Prismatic

Robot experience
      ↓
Semantic event
      ↓
Evidence
      ↓
Individual memory
      ↓
Site knowledge
      ↓
Organizational memory

Again:

Sanctuary asks what the AI should learn to do better.

Prismatic asks what the machine and organization should remember.

That's the important distinction.

8. Sanctuary's hardware-agnostic direction is particularly important

This is perhaps the biggest competitive overlap yet.

Your Prismatic philosophy is:

Different chip. Different OS. Different model. Same cognitive life.

Sanctuary is now saying its Physical AI can operate across multiple robot embodiments, including conventional industrial robots, different mobility options and future humanoids. sanctuary.ai

So both companies believe the intelligence layer shouldn't be tied to one robot body.

The distinction therefore becomes:

Sanctuary portability

Same Physical AI across multiple embodiments.

Prismatic portability

Same persistent identity/memory across multiple embodiments and AI systems.

That is subtle but very important.

9. An excellent test: replace Phoenix with another body

Imagine:

1. Year 1

   Phoenix
   +
   Carbon
   +
   Prismatic identity: Worker-17

   Then Year 3:

   Industrial mobile manipulator
   +
   Carbon vNext
   +
   same Prismatic identity: Worker-17

   The physical embodiment changed.

   Carbon changed.

   But:

   work history
   relationships
   approved procedures
   past incidents
   validated memories
   organizational responsibilities

   remain.

   That would demonstrate something neither the robot body nor behaviour model needs to own.

10. Sanctuary also gives us a Canadian angle

Sanctuary AI is based in Vancouver, and it has received strategic investment from Canadian institutions including BDC Capital and InBC. sanctuary.ai

That makes Phoenix/Carbon useful as a Canadian Physical-AI case study alongside Canadarm3.

But unlike Canadarm3:

• Canadarm3 is a highly specialized flight robotics system;
• Sanctuary is building generalized industrial Physical AI.

That gives Prismatic two very different Canadian examples.

11. Where Sanctuary could become a direct competitor

Of everything we've compared, Sanctuary is one I'd monitor quite carefully.

Why?

Because Carbon already possesses:

reasoning
auditability
goal seeking
natural language
domain integrations
fleet management
hardware independence

Those are relatively high layers of the stack. sanctuary.ai

If Sanctuary adds:

persistent semantic memory
machine identity
relationships
organizational knowledge
cross-model cognitive migration

then it moves directly toward Prismatic's territory.

So this one is both:

validation

and

competitive warning.

13. Prismatic Studio for Sanctuary

I'd model it:

SANCTUARY INDUSTRIAL STUDIO

Orion
Operations Coordinator

Gaia
Organizational Memory

Lexa
Evidence & Governance

Atlas
Maintenance / Reliability

Phoenix-12
Persistent machine identity

Phoenix-17
Persistent machine identity

Then maintain the same hierarchy we've developed:

Machine Memory

What did Phoenix-12 personally experience?

Site Memory

What has Factory A learned?

Fleet Memory

What lessons apply across robots?

Organizational Memory

What procedures and decisions have been validated?

That model works exceptionally well here.

14. Phoenix compared with the others