Memory
# GlassAgent / DocuClear — Foundational Vision v0.1 **Status:** First crystallized version; intentionally open for iteration **Date:** 2026-08-22 ## 1. GlassAgent thesis GlassAgent is an intelligent work-environment architecture. It is not an application, chatbot, or standalone model. It is a three-layer architecture—presentation, business logic, and data—with intelligence built into each layer. The purpose of GlassAgent is to create the conditions in which humans and agents can perform real work together. Work begins with a conversation about what needs to be done. Through iterative work design, the user and an orchestration agent establish the ontology, work objects, contracts, relationships, capabilities, consequences, and required outputs. ## 2. The GlassAgent architecture ### Work design Work design is the initial and ongoing conversation through which the work environment is defined. It creates a shared understanding of the work and establishes the canonical map of the environment. ### Canonical map The runtime maintains a structured map, represented as JSON or an equivalent model, describing the current work surface. It is analogous to an OpenAPI specification, but describes a live intelligent work environment rather than only an API. It defines the ontology, work objects, contracts, workstations, agents, tools, policies, data, business logic, presentation surfaces, relationships, and capabilities. ### Three intelligent layers - **Presentation:** An intelligent, mutable working surface. The UI is data and context, not merely a display. It can be created, inspected, and changed by humans and agents. - **Business logic:** Intelligent orchestration, workstations, contracts, agents, procedures, transformations, calculations, consequence mapping, validation, simulation, and challenge handling. - **Data:** A connected substrate of data items, documents, media, facts, relationships, work objects, calculations, evidence, decisions, provenance, state, and supersession history. ### Workstations and contracts Work moves through purpose-built workstations. A workstation provides the data, context, tools, procedures, agents, interaction surface, and human role required for a defined piece of work. Each workstation operates under a contract specifying what it receives, what it must do, what it must produce, and what evidence is required. The system governs work according to consequences, not merely technical operations. The ontology maps consequences to work objects, properties, relationships, and transformations. The contract manager determines appropriate integrity, approval, simulation, evidence, and challenge rules for the work. ## 3. Persistence, graph, and evolution The entire workspace is a living graph. Canonical nodes are connected through explicit edges such as `derived_from`, `supports`, `depends_on`, `supersedes`, `contradicts`, `affects`, and `validated_by`. Important changes supersede rather than silently overwrite prior state. When a node changes, the graph can be followed to reveal the cascade of affected data, calculations, contracts, workstations, decisions, and outputs. Users and agents can inspect the consequences and decide what to recompute, invalidate, branch, or revalidate. Each interaction is stored. Context is best understood as a query over the current work graph, guided by the user’s language and active workspace surface. The system retrieves the relevant context for each turn rather than reconstructing the entire conversation transcript. ## 4. Human-agent interaction Humans should not be approval bottlenecks. They should be active participants in the work. Agents may continue safely through simulation or mock capabilities when real-world consequences require human input, allowing the user to compare possible outcomes without material side effects. The user can interact conversationally or inspect and manipulate the underlying work environment directly. Different users may see different representations: conversation, tables, graphs, timelines, canvases, forms, or Flow-style executable surfaces. The user and agent can decide together how the work should be visualized. The UI itself is a persistent context and working artifact. It represents the evolving thread of the conversation and may be mutated by both human and AI. Its layout, selections, annotations, relationships, and active areas can communicate work intent. ## 5. DocuClear implementation DocuClear is an implementation of GlassAgent for document- and data-driven work. It exposes persistent GlassAgent work environments through MCP so external agents—ChatGPT, Claude, Perplexity, coding agents, and other compatible systems—can enter the environment and perform work alongside internal agents and the user. DocuClear does not primarily compete with users’ preferred LLMs. It gives those agents a place to work, with the data, context, tools, workstations, contracts, evidence, and persistence they need. ### Product statement **DocuClear gives your preferred AI a place to do real work—with the data, tools, other agents, and human collaboration it needs.** ### Technical statement **DocuClear is an MCP-enabled implementation of GlassAgent: a persistent, contract-bound work environment where external agents collaborate with internal agents and humans to produce inspectable, evidence-bearing work.** ## 6. Short version **GlassAgent is the architecture. DocuClear is the product.** GlassAgent makes presentation, business logic, and data intelligent and coordinated around a defined work object. DocuClear implements that model as an MCP-accessible environment where external agents and users can perform real work together. ## 7. Open questions for iteration - What is the canonical map’s precise schema? - What is the minimum viable work object and contract? - How are external agents authenticated, scoped, and governed through MCP? - How are simulation and real-world side effects separated? - Which parts of the UI are semantic data versus presentation-only state? - How are context queries generated and policy-checked? - How does the system support multiple graph and storage implementations? - What is the first DocuClear product experience that best demonstrates GlassAgent? - Which parts of GlassAgent are proprietary architecture, and which rely on open-source components? ## One-line metaphor **GlassAgent is the intelligent architecture; DocuClear is the place where outside agents come to work.**
Tags: glassagent, docuclear, foundational-vision, architecture, product, mcp, work-design, volatility:state, project:docuclear, duplicate-candidate, kind:semantic — Source: mcp — 2026-08-22 13:22:15 UTC