AI that keeps working when the network does not.
Agencies and government contractors need AI that runs on infrastructure they control, operates without an internet connection, and produces records an inspector general could follow. Data.co Control is built to run disconnected, with every decision made and logged locally.
Sovereignty means the vendor is optional at runtime.
Government data carries handling requirements: controlled unclassified information, law enforcement sensitive material, personal data covered by the Privacy Act, and mission data that may never touch a commercial network. Many AI services assume a constant connection to the vendor for inference, licensing, or telemetry.
Data.co Control makes every access decision inside the deployment. Models run locally on open weights, the policy engine needs no external service, and the audit log is written to storage you control. Licensing is designed to work offline, and updates arrive as signed bundles you import through your own transfer process.
How an air-gapped deployment stays current.
- 1
Stage
Software, model weights, and policy bundles are downloaded on a connected system and verified against published checksums and signatures.
- 2
Transfer
Bundles move across the gap through your approved media and scanning process.
- 3
Verify
The disconnected environment verifies signatures again before anything is installed.
- 4
Operate
Inference, retrieval, policy, and audit all run locally. Nothing calls home.
- 5
Export evidence
Audit records are exported to your records systems on your schedule, never to Data.co.
Mission and back-office uses.
Records search
Search case files and correspondence with results filtered by clearance attributes and need-to-know groups.
FOIA processing support
Draft redaction suggestions and responsive-record summaries for analyst review.
Regulation and guidance Q&A
Answer staff questions from agency policy, regulations, and directives, cited to the controlling text.
Contract file review
Contracting officers check files for required clauses and documents, with procurement-sensitive data restricted.
Designed for government environments.
- Runs fully air-gapped, with no outbound connections required
- Open-weight models served locally; no external inference
- Attribute-based policies for clearance and need-to-know
- Integration with agency identity providers, including smart-card backed SSO
- Signed, verifiable software and policy bundles
- Local append-only audit log with export to agency systems
Common data categories and the control pattern for each.
Categories and markings are defined by your agency and its governing authorities. Data.co carries them as labels and attributes; it does not decide what is controlled.
| Category | Typical handling expectation | Data.co pattern |
|---|---|---|
| Controlled unclassified information | Access limited to authorized holders with a lawful purpose | CUI label, need-to-know groups, local endpoints only |
| Export-controlled technical data | Access limited by citizenship or authorization | Attribute-based deny for unauthorized identities |
| Privacy Act records | Use limited to the purposes in the system of records notice | Purpose conditions on every request |
| Law enforcement sensitive | Need-to-know within named units | Unit groups, export denied |
| Procurement-sensitive | Limited to the acquisition team | Acquisition-team scope, approvals for sharing |
Clearance attributes and purpose, enforced together.
Attributes such as clearance level or program membership come from your identity provider. The rule requires both the attribute and a declared purpose; a missing value is a denial. Illustrative only.
id: pol_gov_ntk_001
description: Program records require membership and a mission purpose
subject: { type: [user, agent] }
resource: { collection: "program.*", classification: [CUI] }
action: [read, search]
when:
resource.attr.program: { in: { ref: subject.attr.programs } }
purpose: [mission, oversight]
effect: allow
otherwise: deny
explain: "Program records require program membership and a mission or oversight purpose."Illustrative example · final API surface documented in /docs
Government authorizations are not yet published.
Data.co has not published any FedRAMP, StateRAMP, or agency authorization. Because Data.co Control runs inside your boundary, it is typically assessed as part of your system authorization. Status updates will be posted on our compliance page.
Questions buyers ask
Does Data.co require internet access?
No. Every runtime function, including policy evaluation, inference, retrieval, and audit logging, runs inside the deployment. Updates are delivered as signed offline bundles.
Is Data.co FedRAMP authorized?
No authorization has been published. Data.co Control is deployed inside your boundary rather than offered as a cloud service, so it is generally assessed as a component of your system.
Which models can run disconnected?
Open-weight models served locally through vLLM or Ollama. Model selection depends on your hardware, mission, and any agency restrictions on model provenance.
Can contractors deploy this for an agency program?
Yes. Systems integrators can deploy and operate Data.co inside a program's environment. See our partners page for how that relationship works.
How is licensing handled without a connection?
Licenses are signed files verified offline against a public key you obtain in advance. Renewal means importing a new license file through your transfer process; nothing calls back to Data.co.
What evidence does the system produce for an assessment?
Policy files with version history and test results, a hash-chained audit log that can be verified offline, configuration files, and checksums for every installed image. These support control assessments for access, audit, and configuration management.
See it against your own use case.
A 30-minute technical walkthrough, policy simulation on your scenarios, and a deployment recommendation for on-prem, private cloud, or hybrid.