Question 1

What is Tentra?

Accepted Answer

Tentra is persistent memory for AI coding agents. It does two things: (1) indexes your repository into a queryable graph of files, symbols, imports, and call edges — agents query the graph via MCP instead of re-grepping source every session; (2) generates visual architecture diagrams from natural-language prompts with 14-framework code export. Integrates with Cursor, Claude Code, Codex, and Windsurf.

Question 2

What is the code graph and why does it save tokens?

Accepted Answer

Tentra indexes your repo into a persistent graph — files, symbols, call edges, imports, semantic purpose. A typical agent session re-reads 20-50 files to build context; a graph query returns the same answer in a single tool call. On our own monorepo we measured a 99.4% token reduction (156.8× ratio) across 8 'where is X implemented?' queries — 114,644 tokens via file re-read vs 731 tokens via graph query. Conservative headline: up to 80% fewer tokens per agent session.

Question 3

Do I need an OpenAI or Anthropic API key to use Tentra?

Accepted Answer

No. Tentra uses the agent-as-LLM pattern — the Claude, Cursor, or Codex agent you already have does the semantic extraction and embedding work on its side. Zero LLM cost on Tentra infra, zero API key setup for end users. Works out of the box with any MCP-compatible agent.

Question 4

How does Tentra compare to IcePanel, Structurizr, and Eraser?

Accepted Answer

Tentra is the only architecture tool that lives inside your IDE via MCP, exports real production code (not just diagrams), and automatically scans codebases to detect architecture drift. IcePanel ($40-80/mo) focuses on C4 model visualization. Structurizr uses a DSL that requires manual coding. Eraser.io generates diagrams from prompts but cannot export code. Tentra is free in beta and offers all these capabilities.

Question 5

What programming languages can Tentra export to?

Accepted Answer

Tentra exports production-ready project scaffolds in 14 formats: Java (Spring Boot 3.2), Node.js (Fastify + TypeScript), Python (FastAPI + SQLAlchemy), Go (chi + pgx), Rust (Axum + tokio), .NET (ASP.NET Core 8), Kotlin (Ktor), PHP (Laravel 11), Ruby (Rails 7 API), Elixir (Phoenix 1.7), plus Docker Compose, Terraform, Mermaid diagrams, and Architecture Decision Records (ADR).

Question 6

How do I set up Tentra with Cursor or Claude Code?

Accepted Answer

Add a .mcp.json file to your project root with the Tentra MCP server configuration. Then describe your architecture in natural language — the AI will call Tentra's tools (create_architecture, analyze_codebase, export_architecture, etc.) to generate and manage your system design. Setup takes less than 60 seconds.

Question 7

Can Tentra scan my existing codebase?

Accepted Answer

Yes. Tentra's analyze_codebase tool scans your project directory and automatically detects services from package.json (Node.js), pom.xml (Java/Maven), go.mod (Go), pyproject.toml/requirements.txt (Python), docker-compose.yml, Kubernetes manifests, and Terraform files. It infers connections between services, detects databases, queues, and external services, then generates a complete architecture diagram.

Question 8

Is Tentra free?

Accepted Answer

Yes, Tentra is currently in free beta. All features are available at no cost including AI architecture generation, 167 cloud components, 14 code export formats, codebase scanning, drift detection, and the full MCP integration. No credit card required.

Question 9

What is the Model Context Protocol (MCP) and how does Tentra use it?

Accepted Answer

MCP (Model Context Protocol) is a standard that allows AI coding assistants like Cursor, Claude Code, Codex, and Windsurf to interact with external tools. Tentra provides 23 MCP tools: 9 architecture tools (create_architecture, update_architecture, get_architecture, list_architectures, analyze_codebase, lint_architecture, sync_architecture, export_architecture, create_flow); 4 code-graph write-path tools (index_code, index_code_continue, record_semantic_node, get_index_job); and 10 code-graph read-path tools (query_symbols, get_symbol_neighbors, get_service_code_graph, explain_code_path, find_similar_code, record_embedding, list_god_nodes, get_quality_hotspots, list_snapshots, diff_snapshots).

Question 10

What is architecture drift detection?

Accepted Answer

Architecture drift detection compares your saved architecture diagram against the current state of your codebase. Tentra's sync_architecture tool scans your code, detects what services exist now, and reports differences: services added in code but missing from the diagram, services in the diagram that no longer exist in code, and changed connections. It provides an accuracy score from 0-100.

Why Your AI Coding Agent Has Amnesia (and How a Persistent Code Graph Fixes It)