Super Intelligence · AI Cloud Foundry

SI Hive
Mind

A network of AI agents that can prove who they are, and prove what they did.

INVESTOR & EXECUTIVE BRIEFING · THOMAS CAPITAL HOLDINGS · NOVEMBER 2026

The story in one line

Agents will do the work.
Trust decides which ones you let in the building.

01

We built the trust layer.

Every agent carries a cryptographic identity, signs what it sends, and leaves a tamper-evident record.

02

We run it on our own metal.

Nine servers, 340 cores and 94 Gb/s of private network, operated as one equal-node hive.

03

We are shipping it in phases.

Identity and trust are delivered. The protocol and hub are built and in testing. Intelligence is next.

The opportunity

The next enterprise question is not what can an agent do?

It is whether you can trust the one that just asked for access. Organisations are moving from one assistant to fleets of agents that talk to each other. The winners will be the ones who can answer three questions on demand.

IDENTITY
Who is this agent?

A certificate issued by our private authority, valid for 24 hours, tied to one registered agent.

AUTHORITY
Is it allowed to do this?

A policy engine decides who may send which kind of message to whom, before anything is delivered.

PROOF
What did it actually do?

A hash-chained ledger and content credentials give every action a record no one can quietly edit.

What we built

TAMP: every message arrives sealed and stamped.

Trusted Agent Messaging Protocol. Think of a registered letter for AI. The sender signs it, the hub checks the seal and the rules, and the receiver verifies it again before acting. A message that cannot prove itself is simply not processed.

Agent Asigns the message
sealed envelope
Hub
  • Verify identity
  • Apply policy
  • Route message
  • Write to ledger
checked delivery
Agent Bverifies the sender
Phase 1 · Delivered
15/15

steps of identity and trust, complete.

The foundation is in place: a private root of trust that every other capability stands on.

    Phase 2 · In testing

    The hub is built.
    Now we prove it.

    All twelve workstreams of the protocol and hub are complete and moving through end-to-end, chaos, load and security testing.

    The hive by the numbers

    Serious capacity, owned outright.

    340
    CPU cores · 612 threads
    984
    GiB of memory
    54.5
    TB raw storage · 56 drives
    94
    Gb/s of linked 10 GbE

    Nine servers in service: five Dell PowerEdge (about 1,040 W measured on those) and the four-node Mac Studio Fleet for GPU compute.

    Hive OS · In build

    A hive that finds itself.

    Power on a node and it announces itself, discovers its peers and joins. Five independent layers make sure it always can.

    1
    Signed report-in roster · in Azure Blob
    2
    Firewall lookup
    3
    mDNS
    4
    Subnet sweep · the fail-safe
    5
    SWIM gossip
    Hive OS · One image

    One image. Every node. Any role.

    THE BASE IMAGE
    Hive daemon
    Latest Claude Code CLI
    Docker, K3s + KEDA
    Same on every server
    +
    SIGNED RELEASE CHANNEL
    Versioned roles

    Pulled from Azure Blob with infrastructure (double) encryption.

    RegistryWebMonitoringLakehouseControl plane
    +
    TRUST BUILT IN
    Identity from the TAMP CA, mutual TLS
    SSH certificate authority in Key Vault
    Auto-registration with monitoring

    An interactive installer turns bare metal into a trusted, monitored member of the hive.

    Operations & CI

    Watched live. Building constantly.

    ● EDIMCSPRD002: 108 agents ● EDIMCSPRD003: 108 agents

    216 self-hosted CI agents serving 54 organisations.

    Mac Studio Fleet nodes EDIMCSPRD002 (M5 Max, 18 cores, 36 GiB) and EDIMCSPRD003 (M4 Max, 14 cores, 36 GiB) run them all.

    CLOUDPORTFOLIO RMM

    Every server monitored live with 21 health checks, full hardware inventory, services, containers and pods, audited remote actions, and CI agent monitoring.

    Proving ground

    A news network comes home to the hive.

    The Meridan Group's news network, MN Current and its sister brands, is moving on-premises. It is the first production workload to prove that the hive can run a real audience.

    ReadersMN Current & sister brands
    Cloudflarein front
    One static addressfront door
    DeltaKubernetes control plane · registry · webKEDA scales pods on demand across Alfa, Bravo, Charles and Echo
    Roadmap

    Two phases down. The rest is on rails.

    1
    COMPLETE

    Foundation & Trust

    2
    IN TESTING

    Protocol & Hub

    3
    NEXT

    Intelligence Layer

    4
    Q4 2026

    TAMP 2.0 Command Layer

    5
    2027

    Enterprise & Compliance

    Next up inside the build: JSON Schema and spec alignment, an A2A binding, hub database migrations, and the nine-server hive on new Hive OS images. Phase 5 targets ISO 27001, 27017, 27018 and 22301 alignment.

    The invitation

    The agent era needs a trust layer. We are building it.

    Identity delivered. Protocol in testing. A hive of our own to run it on. Come see it working, and help shape what comes next.

    WATCH THE FLEET LIVE
    stats.cloudfoundry.ai