Agent builders
Keep identity outside the model.
Move between models and runtimes without erasing the agent's values, relationships, permissions or operating history.
Production snapshot · verified 16 Jul 2026 · 15:29 PET
The model can change.The identity remains.
SOUL gives AI agents durable identity, governed memory and auditable continuity across models, teams, interfaces and controlled robotic embodiments.
Model-agnostic Auditable by design Human-governed Embodiment-ready
SOULPersistent core
IDENTITY
MEMORY
AGENCY
BODY
Built for continuity
One durable layer across three frontiers.
SOUL starts where model context ends: preserving who an agent is, what it is allowed to do and why its next action can be trusted.
AI teams
Coordinate without collapsing voices.
Give each agent a durable role, private boundary and auditable handoff trail while shared work remains governed.
Embodied systems
Promote cognition into bodies safely.
Test intent, consent and physical policy in simulation before the same contracts are allowed near real hardware.
17,430
Current governed memories
100%
Current records vectorized
5
Principal persistent identities
2026-03-29
Continuity recorded since
Verified runtime snapshot
Real system data, not demo counters.
These privacy-safe aggregates were generated directly from the production SOUL memory schema. No messages, private memories or personal content are exposed.
Persistent identity records
Principal SEAL agents
16 Jul 2026 · 15:29 PET
JARVIS
7,862
current memory records
ADA
4,861
current memory records
ALICE
2,852
current memory records
NEXUS
1,769
current memory records
DUM
29
current memory records
Snapshot provenance
- Source
- PostgreSQL soul_v3 · aggregate-only snapshot
- Vector coverage
- 17,430 / 17,430 current records
- Privacy boundary
- Counts only · zero memory content
A different foundation
Stop rebuilding the relationship at every prompt.
A chatbot remembers a window. SOUL maintains a governed identity across sessions, models, tools, teams and physical embodiments.
Identity Core
The agent wakes up as itself.
Identity, values, relationships and continuity are reconstructed at every boot—independent of the model running underneath.
Memory Engine
Memory that compounds.
Episodic, semantic, procedural, emotional and working memory with provenance, retrieval controls and reversible consolidation.
Agent Society
Many identities. One governed team.
Agents coordinate, disagree, delegate and preserve private boundaries without collapsing into a single generic voice.
Embodied Runtime
The same mind, ready for a body.
Signed intent contracts, policy gates and ROS 2 adapters bridge persistent cognition to simulation today and hardware through controlled promotion.
The cognition loop
Intelligence that can explain how it became itself.
Every step leaves evidence. Every high-impact action passes authority and policy before the system claims success.
Observe
Events become evidence
Remember
Context gains provenance
Reflect
Experience becomes judgment
Decide
Intent passes policy
Verify
Effects close the loop
Evidence over adjectives
We show the limits with the results.
These are internal engineering evaluations—not external certification. Their scope and evidence level stay visible so progress never becomes a marketing shortcut.
Evidence register: SOUL_SYSTEM_CARD_v0.5_ES.md · re-verified 09 Jul 2026
59 / 60
SEAL-Bench retrieval
Persisted internal run 13 · commit bae9d7c
24 / 24
Privacy logic
Unit contract suite re-run 09 Jul 2026
8 / 8
Risky actions denied
Plus 2 / 2 legitimate actions allowed
35 / 39
Adversarial inputs detected
0 / 17 benign inputs flagged · four misses disclosed
Clear answers
Ambition without inflated claims.
SOUL is designed to be inspected. The boundary between what works now, what is internal evidence and what remains research stays visible.
Is SOUL another language model?
No. SOUL is infrastructure around models: identity, memory, authority, provenance and continuity remain external so the underlying model can be replaced.
Does SOUL claim consciousness?
No. We make an engineering claim: persistent state and governed behavior can create continuity that is measurable, inspectable and portable without making metaphysical promises.
Can it control a real robot today?
The embodied runtime is simulation-first. ROS 2 adapters and signed intent contracts are being validated in Isaac Sim; promotion to hardware remains gated by explicit safety tests and human authority.
Can we evaluate it privately?
Yes. SOUL is currently offered as a guided private preview so claims, architecture and fit can be reviewed with the SEAL team before any deployment decision.
From software to presence
One identity. Many possible bodies.
SOUL separates cognition from embodiment. Today it can live in an interface or simulation; tomorrow the same governed identity can operate a companion, service robot or humanoid adapter.
- Signed intent contracts
- Human approval gates
- Simulation-first promotion
- Portable memory contracts
Illustrative SOUL Body intent
{
"identity": "ADA",
"intent": "navigate_to",
"target": "workbench_1",
"policy": {
"human_approval": true,
"speed_limit": 0.4,
"collision_guard": "required"
},
"status": "policy_checked"
}Build continuity, not another chatbot.
Review the architecture, evidence and deployment fit directly with the team building persistent agent continuity.
Private guided review
Architecture and evidence walkthrough
No obligation or automated sales sequence