hive.scubed.co · early access

AI under your own keys.

Hive is a sovereign AI system: assistant, coding, and knowledge work AI that can run entirely inside your own infrastructure. Your data never leaves your walls, and your access never depends on a foreign vendor’s policy.

Join the waitlist Hive Enterprise

Frontier AI is becoming an instrument of national policy.

Access is political

Frontier model access has already been suspended and selectively restored by government directive. Rent a foreign model and you inherit that exposure.

The category is proven

Air-gapped government AI now has reference architectures from the largest players. Hive brings that category to ministry and enterprise scale budgets.

Ownership compounds

Self-hosted infrastructure lets you fine-tune and eventually train your own specialized models instead of depending on any external lab indefinitely.

Two ways to run Hive.

Hive

EARLY ACCESS

Hosted by us, open to anyone. The same assistant, coding, and knowledge work capability as the big consumer AI products, billed for what you use.

Assistant, coding agent, and knowledge work in one workspace
Hosted and operated by S Cubed at hive.scubed.co
Usage-based billing: pay for the tokens and credits you use
Live in early access, waitlist open now
Join the waitlist

Hive Enterprise

IN PILOT

For sensitive industries that need control. The full platform installed on infrastructure you own, deeply customizable, and air-gappable when it has to be.

Runs on your servers or private cloud, fully air-gappable
Your keys, your identity systems, your jurisdiction
License-based: includes updates, refreshed models, and maintenance
One-command install on ordinary Ubuntu or Debian servers
Talk to us

Two agents, one engine.

The same category of capability as Claude Code and Codex style tools, running on your models, against your systems.

Coding agent

Reads and writes code, runs builds, tests, and lint tooling, and works with version control against your own codebases and internal tooling.

Knowledge work agent

Drafts and edits long form content, searches your own corpus, generates slide decks, and understands contracts, forms, and scanned PDFs via vision.

RAG on your corpus

Your documents chunked, embedded, and indexed for semantic search, entirely on your infrastructure.

Voice and image

The system reads answers aloud and generates images, all served by local models.

Fine-tuning

Adapt models to your domain on your own hardware, and keep the weights.

Sandboxed execution

Agents run real commands inside rootless containers with no access outside their workspace.

For engineers

Everything inside your boundary.

One compact stack, the same architecture from a single desk side unit to a national scale cluster. In strict sovereign mode, all external egress is blocked at the network policy level for a provable air gap.

APIs
OpenAI and Anthropic compatible API surfaces. Existing tools and integrations point at Hive without modification.
Tools / MCP
A centrally administered tool catalog over the Model Context Protocol connects agents to case management, registries, and internal databases.
Identity
SSO via SAML and OIDC against the identity system you already run. RBAC scoped to departments and users, with feature gating.
Audit
Audit logging with row level security, so every action is attributable, plus cold archive retention for long term records.
Data plane
PostgreSQL with pgvector for retrieval, S3 compatible object storage, all self hosted.
Sandboxing
Rootless server sandbox with per user resource limits and egress allowlists; native OS sandboxing on desktop, same central policy.
Customer controlled infrastructure
Users and apps
Chat and agent UI, desktop agent app, OpenAI and Anthropic API clients
API and orchestration
Authentication, rate limiting, routing, streaming, RBAC, moderation, audit logging
Model routing
Directs each request to configured models, local self hosted models by default
Local models
Tiered LLMs, safety guard model, embeddings, speech, and image generation
Self hosted data plane
PostgreSQL with pgvector, S3 compatible storage, SAML and OIDC single sign on
Strict sovereign mode: all external egress blocked at the network policy layer
Pilot in progress

Hive Enterprise is being piloted with a fintech in Canada and a government agency in Bangladesh.

The problem

Staff were relying on free or personal consumer AI accounts, sending sensitive material to third party clouds that train on conversations by default and can be compelled to produce them.

The deployment

The full Hive Enterprise stack installed on premises on each client’s own servers, running local models with document retrieval over their own records, integrated with their existing single sign on.

The posture

Strict sovereign configuration: external egress blocked at the network policy level, every action logged and attributable, all data inside each client’s own jurisdiction and keys.

Client identities withheld under NDA. Hive is designed to support data residency and organizational control; validate against your own legal and audit requirements.

Run AI you actually own.

Join the Hive waitlist, or talk to us about an Enterprise pilot on one real workflow.