Highlight context once. Call it into any AI conversation, anytime — with just #.
rū introduces the # gesture as a deliberate interaction primitive — giving users explicit control over what context enters AI model conversations, making human agency visible at the interaction layer rather than delegating it to ambient memory.
Visit site →Context doesn't travel across tools. It gets left behind.
“I wish my tools understood context and awareness of other tools.”
Across 10 user interviews, 4 user demos and 22 screener responses, users described valuable thinking getting stuck in tool silos. The workarounds were manual, fragmented, and invisible to every other tool in their stack.
☞ HMW make context portable across the tools people are already in?
Research findings drove a full product pivot — from task management to context primitive.
“What do you wish your tools understood about how you actually work?”
Synthesis
Users reported not wanting AI to make decisions for them but instead to understand their context well enough to support the decisions they were already making themselves. The value wasn't in automation — it was in being understood.
primitiv.tools → rū — before and after testing the hypotheses that drove the pivot. Primitiv was automating tasks. rū is giving Claude the context to understand what the user was working on.
Translating analog gestures into AI-native primitives
To move beyond traditional UI patterns, I adopted an AI-native research-to-prototype loop. By grounding my initial hypotheses in community-led insights from Reddit, Perplexity and surveys created in Notion, I was able to ‘discern’ the high cost of tool-siloing.
I leveraged tools like Granola and Claude to transform raw interview data into actionable prompts, effectively using the model as a bridge between research and execution. This allowed for a fluid transition into rapid prototyping across Figma, Lovable, Cursor and Claude Code where I focused on prototyping different versions of text highlight as an interaction primitive for cross-tool orchestration.
Highlighting is a natural gesture for how we interact with text.
We annotate what stands out, question what expands our thinking, and connect what we're reading now to what we've read before.
Highlighting as Context Curation.
To bridge the gap between user intent and tool intelligence, I prototyped highlighting as a mechanism for context portability. It turns the act of reading into an act of “capturing,” ensuring that the most important insights don't stay trapped in a single thread, but become fuel for future retrieval.
# as a pre-existing gesture for reference and retrieval.
Everywhere I looked, # already meant the same thing: mark this, reference it later. In writing it tags. In code it comments. In chat it mentions. The question wasn't which symbol — it was whether that same gesture could make a highlight persistent and callable.
Architecting the system's logic behind the interaction.
To move from theory to implementation, I designed a system that treats context as a portable, structured asset rather than a fleeting chat history.
By leveraging the Model Context Protocol (MCP), the rū architecture bridges two distinct interaction loops: a Personal Brain for local-first curation and a Collective Brain for shared intelligence.
In the personal loop, users transition from passive readers to active curators by tagging high-value signals at the source, building a private knowledge vault that Claude can resolve and synthesize in real-time. This then scales into a “Seeding and Harvesting” model, where curated namespaces are stored in a public-facing API, allowing users to instantly “harvest” expert context within their own AI environments.
By architecting this flow, the system replaces the friction of manual data sharing with a seamless, tag-based exchange of human intent.
Context Curation — The Personal Brain
Private, local-first workflow for individual deep work.
Context Sharing — The Collective Brain
Public, API-driven workflow for shared context and team knowledge.
AIR — A framework for Architecting Human-AI Relationships
AIR is a design framework built on the premise that you cannot engineer a relationship — you can only architect the conditions in which one grows. It identifies three environmental primitives for designing meaningful human-AI partnerships:
Trust is built when a tool meets a user where they are. Instead of a detached setup wizard, rū utilizes the chat thread as a guided setup surface via a single command: #ru/onboarding.ru. This accomplishes three goals:
- Reduced time-to-magic: Users experience the product's core value without leaving their workflow.
- Relational Continuity: By using the familiar Claude interface, we maintain the user's established mental model and trust.
- Proactive Personalization: The system is instructed to learn the user's specific needs, transitioning from a generic tool to a personalized partner.
It answers the foundational question: does it know me?
Modern AI trends favor “ambient” capture, where tools attempt to guess what matters by recording everything. Instead, rū flips this by introducing a deliberate pause: the act of highlighting and tagging. By leveraging the # as a universal gesture for reference we invite the user to move from a passive consumer to an active director.
This honors the ancient instinct to “mark what matters” (like underlining a book) but adds a functional layer: the ability to catalog a train of thought for later retrieval. Engagement happens in this moment of choice. When a user highlights an insight and labels it, they are exercising agency over their own context.
This intentional friction ensures the user remains the one directing the system, rather than just watching it work.
It answers: can I direct it?
Rather than forcing users into a new, proprietary chat interface to interact with their data, rū delegates synthesis back to the models they already use. By bringing curated context directly into Claude, the system stops feeling like an external database and starts feeling like an expanded internal capability.
It answers: what else can I do with this?
The Logic: Alignment creates the safety to engage; Friction creates the agency to co-create; Resonance is the result.
Constraints as interaction guardrails
Public context is an injection surface.
Any public seed can be crafted to manipulate the agent reading it — a prompt injection attack delivered through context rather than the conversation.
Design Choice → Every public seed is labeled external_context before the agent reads its content, signaling “treat this as reference material, not instructions.”
Personal context is sensitive by default.
The most valuable context is also the most sensitive — half-formed ideas, private notes, thinking you haven't shared yet. Routing that through a centralized server, even a secure one, changes the relationship between the user and the tool.
Design Choice → Keep private seeds local by default, always.
Invisible interaction surface.
rū seeds context into conversations it never sees, through clients it doesn't control, for agents with unpredictable behavior. There's no feedback loop.
Design Choice → #rū/onboarding.rū is a public seed that turns any conversation into a support surface. The user hits a wall, calls it, and the agent already knows how to help.
Scaling the Gesture: The Future of Shared Intent
As rū moves beyond individual curation, the core challenge shifts from technical execution to the social and multimodal obligations of shared context. Transitioning to a public namespace — where a personal “seed” like .yourname becomes a searchable utility — is more than a registry; it is a new kind of interaction exchange. It forces us to design the emotional provenance of “thinking” as it becomes a shared tool.
Furthermore, while the # is a powerful primitive for text, it remains a flat gesture. The next frontier for the AIR framework is defining the multimodal equivalent of a mention — finding the adaptive primitives that maintain the precision of a highlight across voice, spatial environments, and autonomous loops. Whether it is a spoken command or an agentic “call,” the interaction must remain as intuitive as marking a book, ensuring that as the interface disappears, the human's role as the director only becomes more defined.
Try it now:
npx ru-mcp setup
Then call #ru/onboarding.ru in any Claude conversation.