HooRii Introduces HooRii Mind and Open-Sources GCT 0.6B to Advance Personal AI Memory and Retrieval
Hennessy Road, Wan Chai, Hong Kong, Nov. 16, 2026 (GLOBE NEWSWIRE) -- An assistant can understand a request and still
Press Release Disclaimer: This is a press release distributed through the XPR Media network. It has not been independently verified by our newsroom.

![]()
Hennessy Road, Wan Chai, Hong Kong, Nov. 16, 2026 (GLOBE NEWSWIRE) — An assistant can understand a request and still miss the personal history that makes its response useful. In its mid-November 2026 announcement, HooRii Technology is addressing that problem with HooRii Mind (HM), a centralized, visualized memory system for personal context, and the open-source release of GCT 0.6B, its graph context retrieval model. GCT 4B will be offered as a limited HooRii-hosted managed service and will not be open-sourced.
People, relationships, preferences, and events often remain scattered across conversations and applications. Users repeat facts that an assistant has encountered before, or discover that an answer overlooks a family circumstance or an existing commitment. HooRii Mind organizes that background into a graph users can inspect and edit. GCT retrieves the relevant original text from the graph for the AI they choose, connecting organized personal memory to the request at hand.
For HooRii Technology, led by Albert, founder and CEO, the announcement establishes personal context as a foundation for its Personal AI offering. The system gives users a place to maintain their memories and determine how assistants may use them, while the GCT release opens the retrieval method to developer evaluation and deployment.
HM’s Web workspace represents people, their relationships, and the events that connect them through generated memory cards. Users can browse those cards, examine connections, and correct information as circumstances change. A preference can be associated with the person it describes; a shared experience can remain connected to the people involved. Editing works on these organized memories rather than requiring users to rewrite raw chat logs.

The HooRii Mind workspace: a visualized graph of people, relationships, and events, edited as memory cards.
That visibility makes memory management part of the user experience. Someone reviewing the graph can check whether a fact belongs to the right person or whether an event still reflects the current situation. The graph workspace also supports export, giving users a way to take their organized personal context into other settings.
HM separates this workspace from two further capabilities. GCT supplies retrieval for a particular request; graph viewing and editing do not depend on it. Authorized two-way synchronization governs how an accepted agent uses memory and contributes new experiences back to the graph, within the scopes the user grants. A static export provides a copy of information; ongoing synchronization requires its own authorized integration.
Personal graphs are not public by default. Authorization controls who can see which information, so relevance to a request and permission to share remain separate requirements. This allows personal context to serve an assistant within the user’s chosen boundaries, with two-way sync limited to agents whose integrations have passed acceptance and to the access explicitly granted.
Relationships with personas and virtual characters are also part of HM’s memory system. Conversations, shared experiences, and events involving a character can be organized alongside other personal context. HM will support organization and visualization of agent-to-agent, or A-to-A, social links between characters, extending the graph to include those relationships.
Parts of the experience will accept third-party JSON or Markdown materials to build a graph. HooRii has also validated LAN and personal-PC use and will open reference guidance for these configurations. GCT 4B will be offered as a limited HooRii-hosted managed service and will not be open-sourced. The open-source GCT 0.6B release will provide weights, along with deployment dependency references and best-implementation references for developers deploying the 0.6B model. These paths support the broader purpose of making personal context available for inspection, organization, and authorized use.
GCT is graph context retrieval: it retrieves and surfaces relevant personal context from a graph for the AI the user chooses—with lower latency than multi-hop agent tool calls. It does not generate answers.
Chief Scientist of HooRii Technology, describes a method that combines language understanding and graph retrieval within one network. Language layers interpret the request, while graph layers follow relationships to find relevant material. Information found through those connections then informs the model’s continuing interpretation. The relationships help establish whose preference or experience matters, adding context beyond similarity in wording alone.
In a multi-hop agent workflow, following several connections can involve repeatedly generating a tool query, waiting for the result, and reading it before deciding on another query. GCT performs those retrieval steps within the model, reducing the repeated exchanges behind that process. Its output is relevant original text passed to the selected AI; answer generation and task execution remain with that AI and its connected services.

GCT retrieves relevant personal context along graph relations and passes the original text to the AI the user chooses.
A Saturday restaurant request illustrates the value of those connections. A user planning lunch with their mother may need an assistant to account for her preference for lighter food, limited stair access after knee surgery, and the daughter’s piano lesson later that afternoon. Each fact could have appeared in a separate conversation. Following the links among family members, preferences, and events brings the relevant material together for the chosen assistant to consider. The personal agent can then use that context when proposing a venue and time; any booking remains a downstream task.
The mid-November release makes GCT 0.6B weights, deployment dependency references, and best-implementation references available under Apache-2.0 on Hugging Face and GitHub. Developers can evaluate and self-deploy the 0.6B model, examining how it retrieves context from a personal graph for use by another AI. GCT 4B will be offered separately as a limited HooRii-hosted managed service and is not open-sourced.
Together, the workspace and retrieval method give HooRii a way to connect ongoing interaction with usable personal context. For relationships formed with HooRii Workshop personas, HM organizes the events and experiences associated with a character, making that history visible to the user and available for retrieval within authorized integrations.

Relationships and events with Workshop personas, organized in HooRii Mind.
The first hardware application is HooRii’s Stage product line. HM is the cross-generation software memory layer on that same line, from ClawStage (Stage 1.0) to StageGo (Stage 2.0). Best experience is anchored on ClawStage (Stage 1.0) and StageGo (Stage 2.0). HM is not a hardware SKU, and a device purchase does not automatically confer access. Its first seed trial is reserved for ClawStage 1.0 owners, invited in batches.
ClawStage (Stage 1.0) is HooRii’s desktop Personal AI host, giving virtual characters a visible, interactive presence on the desk. It shares Workshop with StageGo for creating and customizing characters. Its completed Kickstarter campaign ran from March 17 to May 1, 2026, with 1,538 backers pledging HK$4,773,656 against a HK$39,150 goal.
StageGo (Stage 2.0), the next generation in the Stage product line, is a portable personal AI communicator planned for launch on Kickstarter. Its one-button interface will offer two modes: Chat for questions, search, and conversation, and Task for delegating more demanding work to a configured AI assistant or agent. With a compatible SIM, StageGo will connect to cloud services over cellular networks without requiring phone pairing. The device’s phone number will also enable supported AI agents to place calls on the user’s behalf, helping complete everyday tasks from start to finish, and help manage text messages, including verification codes. Calling and messaging capabilities will depend on the SIM, service provider, region, and feature availability.

ClawStage, the desktop Personal AI host. StageGo, the portable Personal AI communicator.
Across both devices, HM organizes personal context, including relationships and events with Workshop characters, while GCT retrieves relevant memories for the user’s chosen AI within authorized integrations. GCT does not generate answers or execute tasks. HooRii has validated HM co-location with ClawStage’s Raspberry Pi and will publish reference guidance.
ClawStage 1.0 owners can request invite-only seed access by emailing contact@hoorii.io, indicating ownership and describing their interest in graph organization, GCT retrieval, or authorized memory integration. HooRii will invite owners in batches.

Agnes Wei,Contact@hoorii.io
