Questions buyers ask.

Neural Inverse is available by request. Request access and tell us what you build and where it has to run. We work with your engineering, security and procurement teams from the first conversation. If your organisation already has access, sign in at the address your administrator gave you.

Two. Neural Inverse Build is for engineers: the Editor, the Agents app and the CLI. Neural Inverse Govern, powered by Keystone, is the control plane your organisation runs: sign-in, the models people may use, policy for agents and tools, approvals and the audit record.

The Build apps sign in to your organisation's Govern deployment, and Govern applies your organisation's models, policy and audit to them. The CLI can also run without signing in, with a model provider key or a local model, unless your organisation requires sign-in.

Govern also governs the third-party AI tools your teams already use, as far as each tool allows.

No. Domain modules drive the compilers, simulators, controller connections and host systems you already qualified; the vendor's engine stays yours. The modules are pre-release, and each capability states what it has and has not been run against. See domain modules.

On your own servers, in your own cloud account, or on a network with no internet access. We can also host it for you. The controls and the audit record are the same in each option. See deployment options.

Govern is designed to need no internet access at runtime. It installs from an offline bundle with checksums, and your registry, identity provider, mail relay and model servers sit inside the boundary. Complete offline operation is a property of each release and installation, so run each release on a host with no route out before you rely on it. Vendor audit connectors that poll an outside service can only be used where that service is reachable. See air-gapped installation.

Govern is a single server binary with PostgreSQL 15 or later as its only required dependency. Install it with Docker Compose on one host, or with the Helm chart on Kubernetes 1.29 or later. The Editor and the Agents app run on macOS, Windows and Linux; the CLI on macOS, Linux and Windows. See deploy and operate.

Govern is designed to make no calls to external services at runtime. The Build apps send no telemetry to Neural Inverse, and sign in only to your own Govern deployment. The CLI makes one request of its own: a daily version check against the npm registry, which you can turn off and which is off in CI.

Through Govern's model gateway: Anthropic, OpenAI, Google, Azure OpenAI, Amazon Bedrock and Google Vertex AI, plus self-hosted models on any OpenAI-compatible server or Ollama. Every model is denied until your administrators allow it for the organisation, a team or a project. The Build apps can also use local models, and providers connected with a person's own key where your policy allows. See integrations.

Self-hosted, in your cloud or air-gapped: in your PostgreSQL database and your object storage, on infrastructure you control; we receive none of it. Hosted by us: your content stays in your organisation's home region. On the shared service, sign-in details and billing records are held in the United States; a dedicated deployment keeps them in your region too. Every audit entry records the region that wrote it, and tokens from one region are refused by another.

Yes. Each request is classified from its project, and your routing policy decides which models may serve each classification. By default, restricted work may only reach self-hosted or zero-retention models. Data loss prevention, budgets and audit apply at the gateway whichever client calls it. See model access.

Your policy decides, per action, whether an agent may act, must ask, or is denied. An action that asks waits in Govern's approvals inbox for a named approver. The requester can never approve their own request, every decision needs a reason, and an approval is valid once or for the session, until it expires.

In the Build apps, an approval under an approval rule is a signed record with the signer's name, time, meaning and reason, made after re-authentication and verified in CI. See approvals.

It depends on each tool's own mechanisms, and we state the level per tool. Enforced (blocked before it runs, with the policy in a managed location): Claude Code, Cursor, Windsurf and GitHub Copilot CLI. Managed configuration, with an advisory hook: Gemini CLI. Monitored (recorded, not blocked): Kiro CLI, Cline and Amazon Q Developer. OpenAI Codex CLI is advisory until its enforcement is verified.

Tools that send their model traffic through Govern's gateway also get model access, budgets and data loss prevention. See third-party AI coding tools for every gap.

We assume they could: we design as if an attacker has our source and can run a modified app. That is why anything that must hold is enforced outside it: models, data loss prevention and budgets in Govern, and merge and release gates in your CI on the exact commit. The checks in the apps are a second line. See policy, approvals and audit.

Every change and every enforced decision, written in the same transaction as the change, in hash-chained per-organisation logs with signed checkpoints. Exports carry a signed manifest, and the record can be verified offline with the keys your own installation publishes. See evidence.

No. Govern produces evidence toward the controls of frameworks such as SOC 2, ISO/IEC 27001, ISO/IEC 42001, HIPAA and the EU AI Act. Evidence supports your certification; it is not one. Certification stays with your organisation and its assessor.

Ask us for our current security documentation and questionnaire responses through the contact form. The security architecture in the documentation and our security and trust page are a starting point.

Email security@neuralinverse.com, not the contact form. Our vulnerability disclosure policy sets out what we commit to.

No. Neural Inverse Build and Govern are proprietary, and their source is not published. Under a non-disclosure agreement, your security team can review it. Code the Editor and the Agents app inherit from Microsoft's Code - OSS stays under the MIT License. Neural Inverse OSS, an open-source editor we released in 2026, was a separate, different product; its development has stopped, and its released versions remain under the Apache License 2.0. See licence.

Yes, under your organisation's agreement with Neural Inverse, which sets out the users and uses it covers. Ask legal@neuralinverse.com about licensing.

Once your organisation has access, your administrator provides the installers, together with the address of your Govern deployment. See download and the installation guide.

Start with the documentation; each app's errors end with a reason code your administrator can look up. Existing customers reach us through the contact form with the topic “Support for an existing customer”. See support.