Topology-Aware Development Environment
The end of text-only development.
Built on VS Code. Powered by a Rust Graph Engine.
"Every IDE treats code as a file.
Statuz treats code as a system."
Design Philosophy
Every IDE treats code as text. Statuz treats code as a topological graph. Functions are nodes. Imports are edges. Modules are clusters. Dependencies are relations. This is not a metaphor — it is a computational model, backed by a Rust graph engine with dual-indexed adjacency lists for O(1) traversal in any direction.
Statuz does not replace VS Code — it adds to it. The full VS Code extension ecosystem, debugger, terminal, and language servers remain untouched. The Statuz Layer sits on top, enriching the editor with topological reasoning, AI-powered code understanding, and strategic decision tracking.
A complete topological understanding of any software system can be reduced to three questions. Traverse, Impact, Path. Everything else — health reports, centrality analysis, diff detection, subgraph extraction — is a composition of these three primitives.
The Statuz Manifesto
Core Idea
Three primitive queries. One graph. Infinite compositions.
"What does this connect to?"
O(1) adjacency list lookup. Filter by relation type. Supports cross-field bridge traversal.
traverse(node, relation?)
"If this changes, who is affected?"
Reverse BFS blast radius analysis. Degree centrality. Critical path detection. Cross-field propagation.
impact(node)
"How do I get from A to B?"
BFS shortest path with parent tracking. Field-level path annotation. Cross-field via bidirectional bridges.
path(from, to)
Technical Deep Dive
Written in Rust, compiled via napi-rs, the GraphEngine is the computational heart of Statuz. It maintains an adjacency-list-based directed graph with triple-redundant edge storage for O(1) traversal in both directions. Three copies of every edge exist simultaneously: the global edge registry, the source node's outgoing adjacency (indexed by relation type), and the target node's incoming adjacency (also indexed by relation type). Cascade removal ensures consistency without garbage collection.
Self-contained, encrypted, content-addressable. Magic bytes "STZ\0" identify the format. Content is msgpack-encoded, optionally zstd-compressed (level 3), and ChaCha20-encrypted with an Argon2id-derived key. The Cluster ID is the first 16 bytes of a blake3 hash. The hash covers content only, so metadata can be modified without invalidating it.
A Cluster owns all nodes centrally. Each Field owns its own edges — representing different views of the same ecosystem. Cross-field bridges are bidirectional: each bridge creates forward and reverse edges simultaneously. Cross-field traversal uses recursive DFS with visited-set deduplication. Impact analysis propagates across all fields via reverse BFS, following bridge edges to discover affected nodes in remote fields.
Beyond the three primitives, the engine provides a complete analytical toolkit.
"Three queries. One graph.
Everything else is a composition."
AI Pipeline
A unified provider dispatcher maps 16 providers to their respective chat, FIM, and model-list implementations. Each provider is a first-class entry in a typed record, ensuring compile-time exhaustiveness checking.
for await (const chunk of response)
stream.on('streamEvent', ...)
for await (const chunk of stream)
Two slider types: budget_slider (Anthropic, Gemini) and effort_slider (OpenAI, xAI, OpenRouter). Provider-specific reasoning IO settings. Manual think-tag parsing for open-source models. Per-feature reasoning toggle. Reserved output token space adjustment.
Dual-mode editing: Fast Apply (search/replace blocks) and Slow Apply (full file rewrite). Search/replace has up to 4 retries with block-by-block streaming. Overlap detection prevents conflicting edits. Right-to-left application preserves line numbers. Snapshot-based undo/redo with full state capture. Streaming diff zones with sweep animation.
Custom LRU cache with O(1) lookups, abort-on-eviction for LLM requests. 500ms debounce, 60s timeout. Four completion types: fill-middle, redo-suffix, next-line, no-predict. Stack-based parenthesis balancing prevents syntax errors. Subsequence matching handles LLM-generated suffix variations.
Full Model Context Protocol integration. Dynamic tool discovery, server management, toggle on/off, MCP config file support. Tools are converted to provider-native formats (OpenAI, Anthropic, Gemini) or fall back to XML parsing for providers without native tool support.
Decision Commitment Registry
A formal decision tracking system that registers every architectural commitment, detects drift before it becomes debt, and uses AI to surface implicit decisions the rule engine would miss.
Every decision progresses through a formal commitment hierarchy, from tentative exploration to fully committed. Rejected and superseded decisions are retained for audit trails.
Three scan types: commitment decay (14/7 day thresholds), assumption staleness (21 day threshold), constitution violation (forbidden feature matching, NLP-based constraint detection with Chinese-language support). O(n^2) pairwise contradiction cross-scan across all decisions.
Dependency-injected AI analysis layer. Temperature 0.2 for deterministic outputs. Non-blocking design — AI unavailability never blocks registration. Parses 3 response formats (object, JSON string, markdown-wrapped JSON). Extracts decisions from conversation with confidence >= 0.5 threshold.
Native SVG canvas with 4-layer rendering (edges, nodes, temp-edges, root). Five custom arrow markers. Zoom clamp 200-8000 viewBox. Drag-to-connect with auto-detection. 50-deep snapshot undo/redo. Context menus per node, edge, and canvas. No React, no third-party graph library — pure SVG.
"Every decision is a commitment.
Every drift is detectable."
Capabilities
Context-aware AI assistant with Ctrl+K/L inline editing. Streaming diffs, fast/slow apply modes, 4 retries for search/replace, and autocomplete with LRU caching and parenthesis balancing.
Strategic planning canvas with native SVG rendering. DCR drift detection, 50-step undo/redo, 5 edge types, automatic layout, and drag-to-connect. No third-party dependencies.
ECC catalog browser with one-click install/uninstall. Harness visualization canvas for skills, commands, agents, and rules. Full catalog with 60+ components and 87 commands.
Anthropic, OpenAI, Gemini, Groq, Mistral, Ollama, DeepSeek, OpenRouter, xAI, LM Studio, vLLM, Azure, Vertex AI, Bedrock, liteLLM, OAI Compat. Per-model capability configuration with reasoning support.
Model Context Protocol integration. Full tool discovery, server management, dynamic tool calling. 3 native tool formats + XML fallback. Chat modes: normal, gather, agent.
100% compatibility with the VS Code extension marketplace. 80+ built-in extensions, themes, language servers, and debuggers. Full terminal, Git, and language protocol support.
Why Statuz
Every modern IDE gives you a text editor with AI features. Statuz gives you a topological runtime with AI as a substrate. The difference is not incremental — it is categorical.
| Feature | Statuz | VS Code | Cursor | JetBrains | Zed |
|---|---|---|---|---|---|
| Topological Graph Engine | ✓ Rust-native | ✗ | ✗ | ✗ | ✗ |
| Three Primitive Queries | ✓ O(1) traversal | ✗ | ✗ | ✗ | ✗ |
| AI Providers | ✓ 16 | ~ 2 (GitHub Copilot) | ✓ 8 | ~ 3 | ✓ 6 |
| Decision Commitment Registry | ✓ 6 levels | ✗ | ✗ | ✗ | ✗ |
| Drift Detection Engine | ✓ 5 scan types | ✗ | ✗ | ✗ | ✗ |
| Strategy Canvas | ✓ Native SVG | ✗ | ✗ | ✗ | ✗ |
| MCP Protocol | ✓ Full | ✗ | ✓ Limited | ✗ | ✗ |
| Binary .stz Format | ✓ Encrypted | ✗ | ✗ | ✗ | ✗ |
| VS Code Extension Compatible | ✓ 100% | ✓ Native | ✓ 100% | ✗ | ~ Partial |
| Open Source | ✓ Apache 2.0 | ✓ MIT | ✗ Proprietary | ✗ Proprietary | ✓ GPL |
| ECC Agent Catalog | ✓ 60+ agents | ✗ | ✗ | ✗ | ✗ |
Feature coverage as of Q3 2026. Third-party IDEs may add features over time. Statuz is the only IDE with a native topological graph engine at its core.
Architecture
The Statuz Layer sits on top of the full VS Code Core, preserving all native capabilities. The Rust Graph Engine (in development) provides the topological reasoning foundation. This is additive engineering — nothing is removed, only enriched.
"The text era is over.
The topology era has begun."
Download
Windows 10 / 11, 64-bit
EXE Installer & Portable
macOS 12+, Intel & Apple Silicon
DMG & ZIP
Debian / Ubuntu / Fedora
DEB, RPM, Snap, AppImage
Pre-built binaries are coming soon. In the meantime, build from source.
Development Status