Open-source organizational intelligence

A complete intelligence layer for any LLM.

OrgBox is a free, open-source architecture that packages everything a foundation model needs to be trustworthy in a specific domain — structured knowledge, cross-domain reasoning, behavioral identity, cost-aware routing, interaction patterns, and evaluation — into a single portable artifact. The model is a replaceable utility. The Box is the asset.

Seven layers. One deployable system.

OrgBox is a seven-layer architecture. The intelligence lives in the Box — not in any particular model. Swap the model, the Box keeps working.

L0
Router
Classifies every query by domain, complexity, and stakes. Routes to the cheapest model that can handle it. 85% of queries run locally for free.
L1
Identity
Mission, values, boundaries. Version-controlled, auditable — not hidden in weights.
L2
Reasoning
Twenty domain ontologies connected by a bridge ontology that encodes validated cross-domain relationships. The key innovation.
L3
Knowledge
Curated corpus, vector store, graph store, term bank. RAG with structure.
L4
Tools
Agentic workflows, model adapters, MCP server and client.
L5
Interaction
Structured conversation architectures for real organizational work — mediation, canvassing, intake, facilitation, crisis response.
L6
Evaluation
Sync guardrails, async model-as-judge, compliance layer. The system validates outputs against verifiable criteria — it never trusts its own reasoning text.
L7
Deployment
Docker, Nextcloud ExApp, standalone API, MCP exposure.

Cross-domain reasoning is the contribution.

Knowledge domains are silos. The connections between them — where the same phenomenon appears under different names, where an intervention in one field produces effects in another — are real and consequential, but scattered across literatures that don't cite each other.

The bridge ontology encodes those connections as structured, typed, validated data: concept mappings, structural analogies, causal chains, intervention chains, contradictions, and abstraction mappings. Compressed to ~1–2K tokens and injected into the prompt at query time.

The claim is specific and testable: a small free model with the bridge can match a frontier model without it. We're testing this with a 50-query benchmark. Results will be published regardless of outcome.

Designed around the human, not the model.

OrgBox is grounded in Self-Determination Theory — autonomy, competence, relatedness guide every interaction. The system suggests, never demands. It makes people better at their work, not dependent on the system. It strengthens human connection rather than replacing it.

The organization is treated as a living system, not a machine. OrgBox is connective tissue — routing knowledge, surfacing connections, coordinating workflows — without occupying a position of authority.

20 Human Seats
Twenty documented points where human judgment is irreplaceable — domain expertise, ethical reasoning, cultural context, democratic legitimacy.
Autonomy First
The system serves the person in the seat. It does not occupy it. Machines advise. Humans decide. Always.
Machines advise. Humans decide. Always.

Public infrastructure. Not a product.

Apache 2.0 for code. CC BY-SA 4.0 for specifications and content. No paid tier, no feature gating, no "community edition" missing the good parts. Every organization gets the full system.

Local-First
No telemetry. No training on user data. Data sovereignty is architectural, not policy.
Domain-Agnostic
The first implementation covers 20 social science domains. The template and methodology are published so anyone can build for their field.

Building in the open.

Architecture fully designed — 7 layers, 80+ components. Philosophy frozen. 20-domain knowledge taxonomy defined across 300+ subfields. Sprint plan covers 16 sprints.

Whitepaper Full architecture, bridge methodology, cognitive model, human expertise framework, research agenda.
Section Briefs Every whitepaper section in 75 words and 350 words. The fastest way in.
Diagrams Coming soon
Documentation Component taxonomy, ontology template, sprint plans, build specs.
GitHub Source code, issues, contributions.