Fuzit v1.0 Feature Inventory
Comprehensive capability matrix covering Fuzit v1 across 14 technical domains—from continuous incremental indexing and static AST parsers to graph retrieval, secret redaction, and MCP server interfaces.
1. Repository Discovery & Packaging
Automatic root detection, streaming confinement, gitignore evaluation, and deterministic prompt packaging.
- •Git and non-Git repository root detection
- •Nested monorepo package & workspace boundary detection
- •Canonical repository-relative path normalization
- •Streaming scan engine with token, byte, and file budget limits
- •Repository snapshots and incremental diff deltas
- •Structured output formats: XML, Markdown, JSON, and text
2. Atomic Incremental Index
Content-hash authority database preserving persistent state across edits with zero data loss.
- •Content-free file records with SHA-256 identity authority
- •mtime & size optimization hints for instant change detection
- •Atomic staged index transactions with rollback capability
- •Bounded invalidation fan-out and cycle protection
- •Index verification, corruption recovery, and tombstone tracking
- •Writer locking preventing competing index processes
3. Watch Mode & Local Daemon
Background filesystem watcher ensuring live index freshness without consuming CPU cycles.
- •Native filesystem adapter with repository-root confinement
- •Deterministic debounce and event coalescing
- •Add, change, delete, and directory rename reconciliation
- •Editor atomic-save handling (VS Code, JetBrains, Vim)
- •Overflow detection and dropped-event reconciliation
- •Clean process cancellation and lock contention diagnostics
4. Multi-Language Static Analysis
Zero-execution AST parsers extracting symbols, relationships, and metadata.
- •TypeScript & JavaScript (TSX, JSX, CJS, ESM) parser
- •Python AST module and symbol analysis
- •Java class, package, and interface mapping
- •Go module, package, and replace directive parsing
- •Rust, C/C++, HTML, CSS, JSON, and Markdown support
- •Zero code execution: no compiler plugins or decorators executed
5. Typed Repository Context Graph
Directed node-edge graph mapping repository architecture from symbols to tests.
- •Typed nodes: Repository, Package, File, Symbol, Test, Endpoint, Schema
- •Directed edges: Imports, Exports, Calls, Implements, Tests, References
- •Stable typed node identities with parser provenance tracking
- •Bounded graph traversal with cycle & fan-out protection
- •Incremental graph transactions and tombstone cleanup
- •Safe graph inspection commands via CLI and MCP
6. Graph-Aware Hybrid Retrieval
Task-driven hybrid retrieval combining BM25 keyword matching with AST graph scoring.
- •Lexical task relevance combined with exact symbol matching
- •AST graph distance and edge weight scoring
- •Related test auto-discovery and ranking
- •Git recency, change frequency, and entry-point signals
- •Explicit user anchor protection (--anchor)
- •True --no-index fallback execution mode
7. Explainable Audit & Selection Decisions
Complete transparency into why files and symbols were selected, truncated, or redacted.
- •Per-file score component breakdown (relevance, distance, git)
- •Graph path and expansion distance tracing
- •Budget truncation & omission reason auditing
- •Security redaction justification
- •Zero opaque 'black-box' confidence claims
8. Framework & Ecosystem Intelligence
Evidence-based framework detection for modern full-stack web architectures.
- •Next.js (App Router & Pages Router) route-to-handler mapping
- •React component hierarchy and prop interface tracing
- •Express, Fastify, & NestJS endpoint candidate mapping
- •Python FastAPI, Flask, & Django route detection
- •Spring Boot, Maven, & Gradle Java metadata enrichment
- •Go module replacement & HTTP handler discovery
9. Five Canonical Product Surfaces
Access the exact same repository engine across CLI, Watcher, MCP, VS Code, and Plugins.
- •CLI surface: scanning, packing, graph queries, & doctor checks
- •Local Daemon: live background filesystem reconciliation
- •Read-only stdio MCP Server for AI agents and LLM clients
- •VS Code extension: multi-root workspace trust & inline context
- •Restricted Out-of-Process Plugin Host over stdio
10. Read-Only Model Context Protocol (MCP)
Standardized IPC protocol enabling AI assistants to query code graphs safely.
- •Strict local stdio transport (no network ports exposed)
- •Read-only security boundary (no file write/execute tools)
- •Standard tools: search_context, query_graph, get_file_symbols
- •Seamless connection with Claude Code, Cursor, & Desktop LLMs
- •Schema validation & cancellation-aware requests
11. Local-First Security & Secret Redaction
Strict isolation boundaries protecting private codebases and developer credentials.
- •100% offline local processing with zero telemetry or tracking
- •Built-in secret & credential scanner (AWS, tokens, .env, SSH keys)
- •Sensitive-path filtering & symlink traversal escape prevention
- •Deny-by-default plugin permission isolation
- •Respects .gitignore, .ignore, and custom .fuzitignore rules
12. Cross-Platform Native Engine
Verified native binaries for Windows, macOS, and Linux platforms.
- •Verified on native Windows (CRLF/long paths), Linux (LF), & macOS
- •Case-insensitive filesystem collision safety
- •Locked file handling & clean process cancellation
- •Pre-compiled standalone binaries with zero dependencies
13. Packaging, SBOM & Reproducibility
Enterprise release readiness with verifiable checksums and dependency manifests.
- •SHA-256 checksums for all release binaries
- •Software Bill of Materials (SBOM) json export
- •Clean CI release pipeline with lockfile integrity audit
- •Reproducible build manifests and versioned schemas
14. Evaluation Suite & Quality Benchmarks
Rigorous quality benchmark suite measuring precision, recall, and latency.
- •Polyglot repository benchmark datasets
- •Recall@5 & Precision@5 evaluation harnesses
- •Mean Reciprocal Rank (MRR) tracking
- •Sub-48ms query retrieval latency validation