The Tool Every AI Coder
Will Thank You For
One SQLite file. 26 languages & frameworks. 74 MCP tools with LSP code intelligence. Enable semantic search, relationship mapping, and persistent memory for any AI coding agent—completely local and private.
curl -LO https://downloads.devscriptor.com/installers/devscriptor-1.2.0-x86_64-unknown-linux-gnu.tar.gzOr grab the installer for macOS, Linux & Windows
What Devscriptor Does
for AI-Assisted Development
A toolkit built in Rust for turning your codebase into an AI-navigable knowledge graph
Built in Rust
High-performance, multi-threaded parsing with memory safety and zero-cost abstractions.
Single SQLite Database
All codebase knowledge in one portable file with SQL querying and vector search.
Code Graph Analysis
Entity extraction, relationship mapping, and architectural pattern recognition.
LSP Code Intelligence
38 LSP tools providing IDE features via 4-tier auto-detection system.
26 Languages & Frameworks
Tree-sitter parsing for web, systems, JVM, .NET, mobile, and the Vue & Svelte frameworks.
Gitignore-Aware Discovery
Parallel file walking with native .gitignore support via ripgrep's ignore engine.
Incremental & Real-time
mtime+size fast-path re-scans, plus a watcher that keeps the graph current as you code.
Context Graph Memory
AI agents that remember across sessions with 9 structured note types.
100% Local & Private
Your code never leaves your machine. Zero external dependencies, complete privacy.
MCP Server Built-in
Native integration with 74 total tools: 31 Core + 38 LSP + 5 Architecture.
Semantic Search
Find code by meaning with local Jina embeddings—no API keys required.
Relationship Mapping
Automatic detection of inheritance, calls, imports, and data flow across files.
Install
Download the single universal binary for macOS, Linux, or Windows and add it to your PATH.
Scan
Point Devscriptor at your codebase. It parses 26 languages & frameworks with gitignore-aware discovery.
Analyze
All knowledge stored in SQLite—entities, relationships, and vector embeddings for semantic search.
Connect LSP
4-tier auto-detection finds or installs language servers: Config → Bundled → System → VS Code → Auto-install.
Query
AI connects via MCP with 74 tools: 31 Core + 38 LSP + 5 Architecture tools.
Install in Three Steps
Download the single universal binary for your platform, add it to your PATH, and you're ready. Free to install and use.
Quick Install (Linux x86_64)
curl -LO https://downloads.devscriptor.com/installers/devscriptor-1.2.0-x86_64-unknown-linux-gnu.tar.gz
tar -xzf devscriptor-1.2.0-x86_64-unknown-linux-gnu.tar.gz
sudo mv devscriptor-*/devscriptor /usr/local/bin/Works With Your AI Coding Agents
Native MCP server with 74 total tools: 31 Core + 38 LSP + 5 Architecture
Core Analysis
scan_codebasesemantic_searchget_relationshipsadd_context_notefind_dead_codestart_watchingLSP Intelligence
get_hoverget_definitionget_referencesget_completionsget_call_hierarchyget_diagnosticsArchitecture
lsp_arch_detectlsp_arch_checklsp_arch_batch_checklsp_arch_statuslsp_arch_setupSupported AI Agents
AI Agents That Remember
Long-term memory system that persists knowledge across AI sessions with 9 structured note types
Same SQLite Database
Notes live alongside the code graph in one local file, persisting across sessions and processes.
Linked to Code & Concepts
Attach notes to concrete entities and abstract concepts, then recall them by association later.
Keyword, Semantic & Hybrid Search
Retrieve notes by keyword, vector similarity, or RRF-fused hybrid search with type & priority filters.
9 Structured Note Types
Comprehensive knowledge capture for any scenario
Meet Juggler
Kimi and GLM are among the strongest coding models available — and almost every agent reaches them through a generic OpenAI-compatible shim that quietly drops what makes them good. Juggler is the agent that doesn't.
Preserved Thinking
Reasoning carries across turns instead of being thrown away. Kimi spells it thinking.keep, GLM spells it clear_thinking — with inverted polarity. Juggler translates.
Devscriptor, auto-wired
Detects your Devscriptor install and registers all 74 MCP tools on first run. Zero configuration.
Open weights, on-prem
Kimi and GLM ship open weights. Point --base-url at your own server and the whole stack — model, index, agent — stays inside your perimeter.
Your Code Never Leaves Your Machine
100% local processing. No telemetry. Built for proprietary codebases.
Local-First Analysis
Scanning, indexing, and search make no network calls.
Single SQLite File
Your knowledge graph lives in one portable local file.
Air-gapped Ready
Runs fully offline once the model is cached.
No Telemetry
No analytics, no tracking, no phone-home.