How Pengui AI works

Pengui AI is a self-hosted, white-label enterprise AI platform from Clear Tech. It combines a workspace for users with agent orchestration, company knowledge, memory and business-tool integrations. Your teams share the platform instead of rebuilding that foundation for every AI project.

Evaluate a shared platform when several workflows need common administration, integrations and a consistent user experience, or when deployment ownership and your own brand matter. A single assistant with no integration or governance requirements may not need that additional platform.

In your cloud or data center. The platform deployment includes its interface, engine and storage. Fully disconnected deployments require a separate dependency and licensing review; self-hosting alone does not establish that a system is air-gapped.

That depends on your selected models, connected tools and supporting services. Platform data can be stored in your environment while requests are processed elsewhere. A locally hosted model does not make every integration local. Review endpoints, credentials, logs, retention and network traffic for the complete deployment. Hosted evaluation environments should use non-sensitive data.

Your team controls the deployed platform environment and its access policies. Define operational responsibilities, processing by connected services and any supplier support access in the deployment agreement. Validate these boundaries rather than relying only on the hosting location.

Data in your own deployed database remains on your infrastructure. Software access, export tooling, backup retention and post-termination rights are separate questions governed by the agreement. Agree on and test an exit procedure before production; do not assume every application feature remains available after a license expires.

An agent is a configured assistant for a process, client or workflow. Published plans are sized by agent capacity, with unlimited end users rather than per-user seats. Model usage and infrastructure are separate costs; confirm the applicable plan and implementation scope.

Build around a business task: answering questions over documents, supporting a team process or taking permitted actions in a connected system. Your models, data sources, tool access and workflow design determine what a particular agent can do.

Memory needs the same access boundaries as the data it contains. Configure the intended organization and user scopes, then test authorized and unauthorized access, retention and deletion for your deployment. Memory is a platform capability, not a guarantee that every configuration is safe.

Yes. Professional implementation services can cover discovery, design, integration, testing and deployment. Your team can also build on the platform. Agree on the workflows, integrations, acceptance criteria and ongoing support responsibilities for your project.

Choose supported model providers or compatible endpoints for self-hosted models. Validate tool use, context limits, quality and data processing for your selected model and workflow. MCP connects tools and systems; it is not the model-provider protocol.

Connections use MCP servers or project-specific integrations. Define the systems, authentication methods, permissions and read/write operations you need. Your team or ours can implement integrations; an available API does not make every connection automatic.

Model capability still affects accuracy, tool use and task completion. Retrieval, memory and workflow controls help structure the task, but do not guarantee correct answers or compensate for every model limitation. Evaluate representative tasks, failure cases and the need for human review before changing models.

Deliver the AI experience with your name, visual identity and domain, for internal teams or clients. Confirm the branding scope, organization separation, support responsibilities and any resale rights in your agreement. Branding and deployment ownership are separate from ownership of the platform source code.

The published offer is a 30-day platform trial in your environment, with no credit card required. Budget separately for your model usage, infrastructure and any agreed implementation services. Ask the team about an evaluation sandbox when you are not ready to deploy, and use only non-sensitive data there.

Choose a platform plan to keep running the licensed capabilities you need. Your deployment data and application access are separate: confirm retention, export and license-expiry behavior before the trial ends.