Implementation Layer for Regulated AI Agents
Where humans and agents work together to ensure consistent outcomes, maximise impact, and mitigate risk.
Open products on a shared substrate.
A growing set of production-ready products on a shared typed-memory substrate. Each one composes with the others, shares identity, shares the audit log — and ships AGPL-3.0 from day one, so the community can extend what is here and add what is next.
Substrate
Applications
Connect
Community management — organisations, memberships, federation
Open Connect →BPMN
BPMN 2.0 workflow engine for distributed teams and agent swarms
Open BPMN →Assure
ISO-compliant Quality Management, agentic from the import in
Open Assure →Academy
Course material and training tied to roles and competencies
Open Academy →Events
Events planning, booking, payments, and live gamification
Open Events →Catalog
Product Information Management on the EAV substrate
Open Catalog →Not sure where to start?
Read the architecture overview — substrate, memory, agents, surfaces — and pick the product that fits your stack.
Building memory infrastructure for AI.
Recorded talks and architecture briefings on the substrate behind sovereign agents — covering the OODB working-memory layer, the deterministic commitlog under every reducer call, the cross-LLM portability story, and the AR-glass destination the surface is designed for. Coming soon as the programme rolls out.
The backbone of sovereign agent memory
sovereign twins running on IOBOXX today
typical OODB read latency
deterministic commit log — every reducer call
core licence, sovereign self-host from day one
Built for the AR-glass future.
IOBOXX is an observation surface, not an app. Working memory materialises as floating panels — voice in, glasses out. Today’s spatial simulation runs as a Tauri desktop app where every panel is a real OS window; tomorrow’s primary output is Apple visionOS, Meta Quest, and Google Android XR.
Powering memory for every kind of builder.
Run your agents on a substrate that adapts to your stack and your risk profile.
Sovereign agent infrastructure for regulated work.
IOBOXX nodes run air-gapped or inside your VPC, with deterministic audit trails baked into the substrate. Bring your own models, your own keys, your own grants — and keep a hash-chained commitlog under every reducer call your agents make on your data.
Talk to an architect →For builders & startupsA memory layer you can clone today.
The AGPL-3.0 core is self-hostable from day one. Clone the repo, publish the SpacetimeDB module, point Claude Code or Cursor at the MCP server, and start writing real working memory in under ten minutes — with the same primitives the managed twins run on.
Become a contributor →For platformsEmbed IOBOXX as the memory layer behind your product.
White-label the substrate, expose your own surfaces, keep your own brand. Each downstream twin remains sovereign — its own grants, its own commitlog, its own data residency — while sharing the IOBOXX upgrade cadence and the open MCP contract underneath.
Read the mycelium doctrine →IOBOXX gave us a sovereign brain for each cooperative — same substrate, separate grants, real-time.
Twins running on IOBOXX
Built on open infrastructure
See the latest from IOBOXX.
Updates from the IOBOXX programme.
Item 1 of 3: Memory and model share the same primitive.
The investor whitepaper
A 40-page walk through the substrate (SpacetimeDB-embedded OODB, MCP access layer, Keycloak identity), the memory layer (typed objects, deterministic commitlog, sparse tensors), the contribution economics (Fundi Points → Fundi Tokens on Base), and the IOBOXX mycelium that connects N sovereign twins — written for investors, architects, and regulators alike.
Stand up your own sovereign memory layer.
Clone the AGPL-3.0 core, publish the SpacetimeDB module, point any MCP-aware agent at the MCP server, and watch working memory materialise on the observation surface in real time. Same primitives the managed twins run on, no waitlist, no SaaS sign-up between you and the substrate.
See the architecture
Substrate, memory, agents, surfaces — diagrams and decisions for every layer. Written for engineers joining the platform, with the rationale behind each choice spelled out, not just the result.
Read the docs →Start building
Self-host the core in minutes, wire the MCP server into Claude Code or Cursor with one config block, and your agent is writing real typed working memory on the next reducer call.
Integration paths →