Context-Driven AI Collaboration

Complex creative work requires building knowledge iteratively. Pure conversational tools excel at instant Q&A, but information fragments scatter across chat history, making it difficult to build structure, trace evolution, or reuse across sessions. Writing papers means cycling through reading, thinking, and arguing to gradually construct complete reasoning. Consulting work involves distilling validated analysis methods into processes for reuse in new projects. Product design requires tracing the evolution of decisions to maintain coherent thinking across iterations. This work demands a structured cognitive space.

We built Mooncyan to support this way of working. At its core is Context—a shared cognitive space for you and AI. It holds your information: materials, ideas, analysis, outputs. This space can be iterated on, reused, and shared. Through conversation, you interact with this cognitive space, and AI works from the full context to complete tasks.

Understanding Context

Context is a structured cognitive space. Unlike pure conversational tools where conversations are linear and ephemeral, with information dispersing when the session ends. Unlike document tools augmented with AI features, where documents are static and AI can only passively read.

In Context, you interact with the cognitive space through the natural medium of conversation. You add materials, record ideas, build frameworks. AI works from this structured cognitive space to complete tasks—analyzing, synthesizing, generating.

All information exists as nodes. Documents, ideas, AI's analytical outputs—all are nodes. Relations between nodes show how thinking evolves—where did this idea come from? What analysis supports this conclusion? How did this solution iterate?

This structured organization allows the cognitive space to be iterated on (continuously refined), reused (distilled into templates and workflows), and shared (for team collaboration and knowledge transfer).

When you converse with AI, the structured information in Context—nodes, relations between nodes, accumulated content—helps AI understand the task background and provide more accurate analysis. AI produces analysis as nodes, which you can organize, reuse, or further refine.

This means you can pause a project for weeks and return to find everything still there. More importantly, this preserves the complete working context—you can see the full picture and continue whenever you're ready.

How It Works

Let's understand through a concrete example: you're writing a review paper on machine learning applications in healthcare.

You start by creating a context named "ML in Healthcare Review." Then you upload 15 reference papers—each becomes a node. You define an analytical framework node: "Analyze each paper across three dimensions: methodology, application scenarios, and technical limitations." Next, you ask AI to analyze the papers following this framework. AI extracts nodes for methodology summaries, application classifications, and more. You review AI's analysis, find two papers incorrectly classified, directly edit the node content, and add annotations explaining your judgment.

After several rounds of conversation, the context now contains over 20 nodes. You begin conceptualizing the review framework and write down an initial outline node. Simultaneously, you ask AI to deeply analyze technical details of several key papers—you work in parallel. After AI generates supplementary analysis, you adjust the outline based on this new information and give AI feedback: "The challenges discussion needs to focus more on data privacy issues." AI regenerates the analytical framework for that section based on your direction.

You start writing the introduction following the outline. When finished, you ask AI to evaluate: "Is the argumentative logic of this introduction clear?" AI points out two areas for improvement. You adopt one suggestion but keep the other as is. Next, you write some rough thoughts on methodological evolution—rather scattered. You ask AI to organize this into complete paragraphs. AI supplements the argumentative structure and transitions. You refine the wording and finalize it as a new section node. You can choose to preserve each iteration, tracing the evolution from first draft to final version.

Weeks later, the paper is complete. But Context's value extends beyond this. You ask AI to review the entire context: "Summarize the effective methods and processes from this review project." AI analyzes all nodes and conversation history, distilling a methodology: define analytical dimensions, categorize and extract insights, work in parallel, iterate continuously. You review this summary, adding details AI missed (like handling classification conflicts), removing less generalizable parts. Together, you form a "Review Methodology Summary" node. Next time you work on a similar project, you don't start from scratch—you start from this validated workflow.

This is how Context works: not one-off Q&A, but continuous accumulation. Information isn't lost, experience compounds, methods can be reused.

Beyond Knowledge Base: From Information to Action

By default, nodes in Context are containers of information—documents, ideas, analysis results. AI retrieves information from them as background and basis for tasks. This resembles traditional knowledge bases: you save information, AI reads information.

But you can establish shared understanding with AI about the meaning of certain node types. For instance, you mark a text node as "workflow" and attach a definition: this type of node describes a series of steps, each specifying what to do and how to handle it. When AI encounters such nodes, it knows this isn't just plain text, but a process to understand and execute. This is cognitive alignment between you and AI—reaching mutual understanding about the meaning and handling of these node types.

An example: generating a weekly HackerNews digest. You create 3 nodes: fetch content (code node), synthesize summary (text node), send email (code node), chain them together with relations, and mark each node's role as "workflow." AI sees this structure, recognizes they form a workflow, and executes each step in the order defined by relations. This isn't hardcoding every detail, but providing structure and intent for AI to understand and complete.

Code nodes enable AI to interact with the external world: calling APIs to fetch data, processing files, sending notifications, generating reports. These capabilities allow AI to not only analyze and generate content, but truly execute tasks—from information space to action space.