Subspace

Product Introduction

1. Definition: Subspace is a specialized Agent Workspace and orchestration platform designed specifically for the era of Agentic Engineering. It functions as a unified desktop environment—currently optimized for macOS—that integrates multiple large language model (LLM) agents, terminal emulators, documentation viewers, file systems, and web browsers into a single, cohesive workflow.

2. Core Value Proposition: Subspace exists to eliminate "agent amnesia" and the friction of tool fragmentation in AI-assisted development. By providing a persistent, cross-agent memory layer and a keyboard-first interface, it allows developers to run diverse CLI agents like Claude Code, Codex, and Gemini side-by-side. The primary objective is to transform AI agents from isolated, stateless tools into long-term collaborators that maintain project context across multiple sessions and varied toolsets.

Main Features

1. Subspace Memory Engine: This background system automatically captures and synthesizes every interaction within a workspace. Instead of losing context when a session is cleared (via /clear), Subspace compresses conversation turns into token-efficient observations utilizing structured tags such as decisions, blockers, and progress updates. Every few turns, it generates a status update to synthesize recent work. This memory is shared across all agents in the workspace, meaning context from a Claude Code session is immediately available to Codex or OpenCode, ensuring a continuous "Ship Loop."

2. Multi-Agent Orchestration Environment: Subspace allows for the simultaneous execution of various CLI agents within a single interface. Users can deploy Claude Code, Codex, OpenCode, Gemini, Cursor, Copilot, Amp, Droid, OpenClaw, and Kiro. This feature enables "tag-teaming" agents; if one model reaches a reasoning dead-end, another can be activated with the full context of the previous agent's attempts and failures already loaded into the workspace memory.

3. Integrated Agentic Workspace: The UI is engineered for high-performance engineering, featuring a 100ms workspace switcher. It consolidates terminals, code files, and a live browser into a unified pane. The workspace persists across restarts, maintaining the state of all active agents and open documents. It employs a keyboard-first design philosophy, centered around a Command Palette (Cmd+K) that handles everything from launching agents with specific prompts to searching browser history and executing shell commands.

4. React-Aware Browser Integration: The built-in browser is not a standard web view but a specialized tool for UI debugging. By tapping into the React fiber tree, Subspace allows users to select elements on a page and leave comments that include the exact source file, component name, and line number. When an agent receives this feedback, it doesn't need to "grep" the codebase; it is provided with the precise coordinate of the code responsible for the rendered output.

Problems Solved

1. Context Fragmentation and Re-explanation: Traditional AI chat interfaces suffer from state loss. Developers often spend significant time re-explaining project architecture, previous decisions, and current blockers to agents. Subspace solves this by making memory a property of the workspace rather than the individual chat session or tool.

2. The "Alt-Tab" Cognitive Load: Agentic engineering typically requires developers to juggle terminal windows, IDEs, browser tabs, and AI chat windows. Subspace reduces this complexity by grouping all necessary components into a single workspace, providing better visibility and reducing the mental overhead of switching contexts.

3. Target Audience: The platform is built for Software Engineers, particularly those specializing in "Agentic Engineering," React Developers who require tight integration between UI and source code, and Technical Leads who need to orchestrate complex tasks across multiple AI models.

4. Use Cases: Subspace is essential for complex refactoring tasks where multiple agents must collaborate, rapid prototyping where UI feedback needs to be piped directly to a coding agent, and long-running development projects where maintaining a "decision log" is critical for agent performance.

Unique Advantages

1. Differentiation: Unlike standard IDE extensions or standalone AI chat apps, Subspace is a standalone "command center." It does not lock the user into a single model or vendor; instead, it provides the infrastructure to run any CLI-based agent. Its unique advantage lies in its "cross-pollination" of memory, where data from different agents is synthesized into a single source of truth.

2. Key Innovation: The most significant technical innovation is the automated observation and synthesis layer. By background-processing conversations into structured, searchable memory (searchable by text, tag, workspace, or timeline), Subspace overcomes the context window limitations of modern LLMs, providing agents with "infinite" project history in a token-efficient manner.

Frequently Asked Questions (FAQ)

1. How does Subspace Memory differ from standard LLM context windows? While a standard context window is limited by token counts and resets every session, Subspace Memory works in the background to compress and tag every interaction. It creates a persistent, structured record of decisions and progress that is injected into new sessions, ensuring the agent is always up-to-date without consuming unnecessary tokens on redundant history.

2. Which AI agents are compatible with Subspace? Subspace supports a wide array of CLI-based agents including Claude Code, Codex, OpenCode, Gemini, Cursor, Copilot, Amp, Factory, Droid, OpenClaw, and Kiro. It is designed to be model-agnostic, allowing developers to use the best tool for any specific sub-task within a project.

3. What makes the Subspace browser better for React development? The Subspace browser is integrated directly with the workspace memory and the React fiber tree. This allows it to trace rendered UI elements directly back to their source code file and line number. When you leave a comment on a UI element, the agent receives a technically precise instruction rather than a vague description of the visual bug.

4. Is Subspace available for Windows or Linux? Currently, Subspace is optimized for macOS (Apple Silicon). However, Windows and Linux versions are officially in development to support the broader engineering community.