Stop Feeding Raw Text to Coding Agents.
Give Them Architecture Intelligence.
An embedded, deterministic Code Graph Engine (Graph-RAG) for AI agents. Slashes LLM token bills by 90% and eliminates hallucinations using .NET 10 Native AOT, Kùzu DB, and Tree-sitter over MCP.
Core Engine Architecture
The Zero-Config Monolith
No Python runtimes, no Docker overhead, no Redis, no Neo4j, and no hosted cloud vector stores. Everything runs embedded in-process on bare metal.
.NET 10 Native AOT
Precompiled ahead-of-time into a single native binary. Starts in under 15ms with zero JIT compilation, zero dynamic reflection, and ultra-lean resident memory.
Embedded Kùzu Graph DB
Embedded C++ graph engine via direct P/Invoke bindings. Executes microsecond Cypher queries across topological edges (CALLS, IMPLEMENTS, CONTAINS) with zero external servers.
Polyglot Tree-sitter AST
Deterministic C grammars queried via .scm S-expressions. Indexes C#, TypeScript, JavaScript, Python, PHP, and Markdown without sending any code to an external LLM.
Local ONNX 384D Vectors
Embedded INT8-quantized all-MiniLM-L6-v2 running locally on CPU. Accelerated with hardware SIMD (TensorPrimitives) for instant semantic symbol retrieval.
Model Context Protocol (MCP)
First-class JSON-RPC implementation over stdio using System.Text.Json source generation. Connects out-of-the-box with Claude Desktop, Cursor, Antigravity, Copilot, and Codex.
Continuous Post-Edit Sync
Incremental graph synchronization after every code edit or git commit. Enforces agent rules so your AI assistant never operates on stale architectural knowledge.
Efficiency & Precision
Plain Grep vs. ArchiMap Graph
Traditional AI coding agents read full 1,000+ line files into the prompt, burning tens of thousands of tokens and hallucinating on dead methods. ArchiMap delivers surgical structural slices.
Token Consumption per Typical Refactoring Task
| Capability & Behavior | Vanilla Agent / Raw Grep | ArchiMap Graph-RAG |
|---|---|---|
| Ingestion Scope | Dumps entire files (1,200+ lines, hundreds of unrelated helper methods). | Pinpoints entry point via 384D ONNX vector + 1-hop traversal in graph topology. |
| Token Cost / Round-Trip | ~73,220 tokens ($0.366) per refactoring task. | ~2,180 tokens ($0.011) (surgical AST slice). |
| Context Precision | "Lost in the middle", hallucinates methods, confuses overloads. | Strict deterministic pruned AST: LLM only inspects target slice. |
| Blast Radius Analysis | Blind edits. Requires full build failures to find broken callers. | Calculates upstream callers, implementors & affected tests in <5ms. |
| Team Monthly Bill (15 Devs) | ~$4,393.20 / mo | ~$130.80 / mo (Saves $4,262.40/mo, >850M tokens) |
Enterprise ROI Projections (Amazon Bedrock)
Based on Claude Opus 4.8 benchmark pricing ($5.00 / 1M input tokens) across 40 daily agent prompts/dev
Model Context Protocol
Deterministic MCP Primitives
Exposed over JSON-RPC on standard input/output. Your agent invokes structured graph operations instead of grep scans.
Semantic Entry Point Locator
Uses local 384D ONNX vectors + lexical BM25 matching to return precise symbol node IDs and source file paths.
→ Symbol #4812 in Storage.cs:142
AST Structural Skeleton
Traverses Kùzu DB graph relations (1-hop or n-hop) and returns method signatures and contracts with zero body bloat.
→ 380 tokens (vs 14,200 full file)
Blast Radius & Test Traversal
Computes upstream callers, affected downstream implementors, and targets tests that need verification before edits.
→ 4 Callers, 2 Tests affected
Targeted Slice Inspection
Retrieves the exact implementation body only for the targeted symbol slice, with path traversal safeguards.
→ Exact 35-line method slice
Interface & Base Traversals
Directly queries the Kùzu graph for all concrete types and classes satisfying an interface or abstract contract.
→ KuzuGraphStorage in Storage/
Architectural Pattern Discovery
Discovers recurring architectural patterns (co-occurrence clusters, naming templates, base interfaces, DI registration).
→ DI wiring + companion tests
Reverse Call Hierarchy
Queries upstream graph relationships (<-[:CALLS]-) to surface all symbols invoking a target method or constructor.
→ McpServer, QueryHandler
Macro Topology & Packages
Queries project references, workspace packages (NuGet, npm, PyPI), and target frameworks without touching config files.
→ 3 Projects, 5 Packages
Pattern Completeness Guard
Validates modifications against known patterns, detecting missing companion tests, DI bindings, or interface implementations.
→ Warning: Missing UserHandlerTests.cs
Incremental Graph Sync
In-process synchronization refreshing Kùzu DB graph with Git state after commits or bulk edits, without server restarts.
→ 3 modified files indexed in 32ms
GrandNode 2 Benchmark (.NET Monolith)
Stress-tested against GrandNode 2 — an enterprise open-source e-commerce solution with 57 projects, Clean Architecture, MediatR CQRS, and hundreds of domain models.
Modern Quickstart Guide
Zero-Config Agent Integration
No manual JSON editing required. A single command detects your AI editor and wires the Model Context Protocol automatically.
Install Single Binary
Auto-detected
Fetches latest release (v0.1.0) directly to ~/.local/bin/archimap with companion tools.
Ensure ~/.local/bin is in your $PATH.
Initialize & Index
Initializes local Kùzu DB storage and generates agent awareness files (.cursorrules, GEMINI.md, CLAUDE.md, AGENTS.md).
cd your-project && archimap init && archimap index .
Installs automatic Git post-commit graph sync hook.
Auto-Wire AI Agents
Automatically detects and configures all installed AI agents with a single CLI command.
archimap setup --all
Flags: --cursor, --claude, --antigravity, --copilot
Manual MCP Configuration
Direct file schemas if you prefer manual configuration over archimap setup