Sociail · Backbone
Architecture spine
How shared AI work stays coherent

One coherent backbone for shared AI work.

Sociail keeps rooms, context, artifacts, approvals, receipts, and bounded AI help on one coherent spine so the product can stay useful without becoming opaque.

Visible AI stateReview before boundary crossingsMemory and grounding honestyArtifacts that stay with the roomReceipts for important outcomesSupport handoff when work gets sensitive
See trust commitmentsSee product surfaces
01Core model

The foundation is a single spine that every feature reuses.

Human+Room+AI Teammate+Context+Policy+Memory+Tooling+Approval+Receipt+Work Graph
The Backbone equation

Every major feature has to reuse that spine. If a feature needs its own identity model, permission model, receipt model, memory model, or claim language, that is a warning sign.

02AI participation traits

The five teammate traits map to concrete backbone layers.

Brain, Soul, Memory, Hands, and Heartbeat are not just product language. The sequence moves from context and role through continuity, bounded action, and governed follow-through.
T01Brain
Cortex / contextCortex · Context Firewall

Intent, risk, provider choice, and room context shape what the AI teammate can use before it answers.

T02Soul
SOUL / role doctrineSOUL · Persona · Directive Stack

Identity, role, tone, operating doctrine, and limits define what the participant is supposed to be in the room.

T03Memory
Memory + Data FabricMemory · Data Fabric

Retained context, citations, corrections, and forget requests stay structured and scoped instead of becoming invisible memory.

T04Hands
Tool Broker / actionsTool Broker · Policy + Grants

Actions flow through capability handles, policy checks, approvals, and receipts instead of raw model access.

T05Heartbeat
Follow-throughOperational Spine · Work Graph

Follow-through attaches to tasks, commitments, support state, and room history so resurfacing work is inspectable.

03The Backbone

The delivery backbone uses eight reusable control layers.

L01

SOUL / Persona / Directive Stack

What the AI is, how it behaves, and what boundaries it respects.

L02

Cortex / AI Shield / Model Routing

How intent, risk, provider choice, prompt safety, autonomy, effort, and escalation are routed.

L03

Context Firewall

What context may move between rooms, users, AI teammates, tools, memories, and models.

L04

Policy + Grants + Entitlements

What is allowed, blocked, staged, approved, denied, metered, or upgraded.

L05

Memory + Data Fabric

What is remembered, structured, corrected, forgotten, or cited.

L06

Tool Broker / API Services / Runtime

How tools, service accounts, webhooks, connectors, and runtime actions happen outside model context.

L07

Receipts / Trust Grammar

What happened, why, under whose authority, with what proof.

L08

Operational Spine / Work Graph

Tasks, decisions, commitments, artifacts, budgets, support, incidents, records, and outcomes.

04How features fit

Every surface attaches to the same spine.

The feature portal can show what is ready, in preview, in pilot, or planned. The backbone explains why those surfaces compose safely instead of becoming separate control systems.

First wedge · primary surface

Ready

Shared AI Rooms

The first wedge and primary delivery surface, the room is where the spine becomes visible to the user.

View Shared AI Rooms

Visible AI teammates

Ready

AI Teammates / Participant Agents

Visible AI teammates in shared rooms. User-created teammate templates and deeper team setup remain proof-gated.

View AI Teammates / Participant Agents

Identity

Preview

Sociail Me

User-owned memory, boundaries, style, and eventual alter-ego projection.

View Sociail Me

Knowledge

Pilot

Data Fabric / Memory Palace

Structured room knowledge, review-first and source-linked.

Capability

Planned

Vault

Protected capability use without exposing secrets to models.

Comms

Pilot

Communication Activity

Governed communication activity and endpoints, not raw provider access.

Video + voice

Planned

AI in Meetings

Visible meeting presence, transcript lineage, recap drafts, and disclosure-first controls.

Coordination

Planned

Agent Huddles · AITL · HITL

Agents can coordinate; humans still consent and commit.

Programmatic access

Pilot

API Endpoint Services

Service-account access to governed AI service surfaces without bypassing scope, grants, denials, or receipts.

Platform layer · planned

Planned

Marketplace · Wallet · Trust Anchors

Planned platform layers, only after identity, grants, receipts, disputes, and governance mature.

05Sequencing discipline

The vision is broad. Execution stays narrow.

Now
  • Login-to-signoff MVE
  • Shared AI Rooms
  • AI teammates in rooms
  • Trust-visible AI turns
  • Artifacts
  • Approvals
  • Receipts
  • Memory honesty
  • Support handoff
Proof lanes
  • Room Data alpha
  • Communication activity
  • Managed endpoints
  • Prompt / Cortex metadata
  • Model routing calibration
  • Agent templates / TEAM setup
  • API denial proof
Planned
  • Sociail Me depth
  • Vault handles
  • AI in meetings
  • Agent huddles
  • Richer Data Fabric
Earned, not assumed
  • Public marketplace
  • Broad autonomous purchasing
  • Public AI teammate network
  • Blockchain anchoring as a headline
Most important principle
The backbone is how we prevent ambition from becoming chaos.

The vision is broad, but execution stays narrow: prove one repeated human + AI workflow, then let each future surface attach to the same backbone spine, only when the underlying trust, policy, memory, receipt, and support layers are strong enough.

How Early Access works
Public label: Backbone. AIX is the architecture underneath the label.
Read the trust commitmentsSee the backbone in workflows