SYLVA: A Perception Engine Rendered as a Living Forest

The Problem

As AI agents become more capable and autonomous, developers are transitioning from writing code to orchestrating agents. Current tools — VS Code, terminal multiplexers, chat interfaces — were designed for single-developer, single-project workflows. They completely break under multi-agent, multi-project orchestration.

The core pain: "Figuring out what is where takes more effort than the building part does."

Developers managing 3-5 concurrent projects with multiple AI agents spend more cognitive load on context management than on actual engineering decisions. The IDE, the terminal, the browser, the agent logs, the git history — the developer's attention is fractured across a dozen surfaces, none of which talk to each other.

The Thesis

"The farmer doesn't need a dashboard. They need eyes adapted to their landscape."

SYLVA inverts the IDE hierarchy. Instead of:

OS → Window Manager → IDE → Project → Files → Code

SYLVA proposes:

OS → SYLVA → World (all projects) → Project (organism) → Module (branch) → File (leaf) → Code (cell)

The developer's default mental state shifts from "I am inside this project looking at files" to "I am surveying my entire working universe and can descend into any part of it." This is not a cosmetic change — it restructures how multi-project, multi-agent work flows.

The Architecture: Three Rendering Layers

Living World (Three.js, 2.5D Spatial) Projects are trees. Files are leaves. Dependencies are roots. Agents are luminous entities moving through branches. Health, churn, complexity — all perceivable at a glance through color, size, animation, and spatial position.

Command Surface (React Flow, 2D Operational) Structured task management, approval queues, agent control panels. Information-dense when you need to make decisions, invisible when you don't.

Microscope (Monaco Editor) A world-class text editor — but as one zoom level inside the spatial world, not the outermost container. You descend into code; you ascend back to the forest.

The Antigravity Layer: Beyond the Skin

A spatial UI on top of traditional Git and POSIX filesystems would be a lie. SYLVA's nervous system requires:

Continuous Semantic Knowledge Graph — Real-time AST-based graph database, not flat file grep. When an agent proposes a change, the IDE instantly computes the downstream impact across the entire codebase.
Ephemeral MicroVMs — Millisecond sandbox environments (Firecracker/E2B) for agent hypothesis testing. No shared filesystem pollution. The developer sees 50 experiments compressed into one high-confidence result card.
Blast Radius Compression — Before any merge, SYLVA computes syntactic (files changed), semantic (functions affected by shared state), and infrastructural (env vars, schemas, payloads) impact. The developer sees the highest-leverage decision, not raw diffs.
Agent Cognition Visibility — Expose what agents think they understand about the codebase. Show confidence levels, evidence trails, and scope boundaries as first-class UI elements.

Five Primary Views

World Map — All projects as a living landscape. Morning triage surface.
Canopy View — Single-project architecture. Module health, coupling, churn heatmaps.
Agent Theater — Active agents with scope visualization, overlap detection, evidence trails.
Diff Corridor — Spatial diff review with semantic blast radius.
Merge Gate — Pre-merge risk analysis, conflict prevention, approval workflow.

Why Now

Three converging trends make SYLVA viable:

MCP standardization — Model Context Protocol provides a universal agent orchestration layer. SYLVA speaks MCP natively.
Agent maturity — Claude Code, Cursor, Copilot Workspace — agents are production-ready but tooling hasn't caught up.
WebGPU availability — Browser-native GPU acceleration makes real-time spatial rendering of large codebases feasible.

The Name

Sylva — Latin: (1) a forest or woodland; (2) a descriptive catalog or survey of the trees in a particular region.

It is the only name that captures both halves of the product identity: the living forest AND the perception engine that surveys it. John Evelyn published Sylva, or A Discourse of Forest-Trees in 1664 — a comprehensive catalog that helped landowners see and understand their forests. That is literally what this product does for codebases.

Sylva: See your forest.