High-Performance Native AI Coding Runtime

Documentation & Architecture

Built from the ground up in native Rust

Unlike heavy wrappers and Python scripts, Simplicio compiles down to a single self-contained binary. No virtual environments, no cold-start latency, and zero dependency conflicts on your host system.

Rust Tokio MCP SHA-256

Runtime Engine 100% Rust

Neural Cache Lookup: 177 ns Token Waste: 0 Tokens

01.1 · VANTAGENS ARQUITETURAIS · BENCHMARK REAL

🏆 Nossos Diferenciais Competitivos

O Simplicio não é um mero compressor de texto ou proxy de terminal: ele é um kernel autônomo local-first em Rust que resolve os gargalos estruturais de custo, latência e estabilidade da engenharia de software com IA.

⚡

Prompt Caching > 90% Comprovado Injeção determinística do mapa do repositório no prefixo do prompt (Context Anchor). Em testes com OpenRouter/DeepSeek, alcançou 93,06% de cache hit rate e redução de -64,3% no custo por turno.

🔬

Edição Cirúrgica Atômica (Zero Truncamento) Modifica apenas as linhas necessárias via simplicio_edit. O modelo nunca reescreve arquivos inteiros, eliminando 100% dos riscos de // ... existing code ... e economizando mais de 95% em tokens de saída.

🧠

Daemon Residente Tokio com RAM < 1ms Processo em segundo plano via socket Unix (~/.simplicio/daemon.sock). ASTs, símbolos e grafos ficam permanentemente aquecidos na RAM compartilhada entre múltiplos harnesses concorrentes, com watcher inotify reativo.

🔌

Interoperabilidade com 29 Harnesses Conexão transparente com Pi, Cursor, Claude Code, Codex, Copilot, Antigravity, OpenCode, Devin, Goose e outros 20 harnesses. Inclui auto-detecção de conflito (mapper-first) e unlock dinâmico via tools/list_changed.

🛡️

Schema Resilience & Coerção Segura Camada pré-validação tolerante a variações de LLMs: aceita path por file, cmd por command, ops por operations, sem falhas de validação JSON-RPC.

📜

Recibos Auditáveis de Economia (ROI) Gera comprovantes canônicos (simplicio.token-savings-receipt/v1) com métricas reais reportadas pelos provedores: taxa de cache, tokens evitados e dólares economizados.

// Telemetria Real Observada (Modelo: deepseek-v4.1-flash via OpenRouter):
Turno 1 (Mapeamento Inicial & Context Anchor):  Input: 2.926 | Output: 78 | Cache: 0     | Custo: $0.0024675
Turno 2 (Edição Cirúrgica via simplicio_edit):    Input: 210   | Output: 86 | Cache: 2.816 | Custo: $0.0008809
=> TAXA DE PROMPT CACHE HIT: 93,06%  |  ECONOMIA DE CUSTO NO TURNO: -64,3%

Installation

One command to get started

Install the signed, standalone binary on your platform:

curl -fsSL https://simpleti.com.br/simplicio/install.sh | sh

irm https://simpleti.com.br/simplicio/install.ps1 | iex

pip install --upgrade simplicio

Verifies the checksum and installs the executable for your platform.

Lifecycle

End-to-End Request Lifecycle

How Simplicio coordinates every user request from prompt ingestion to verified delivery:

1. MCP Protocol Ingestion

Your AI agent (Claude, Cursor, Antigravity, etc.) sends tool invocations via standard JSON-RPC 2.0 without leaving the chat.

2. CPUExecutor Lane Classification

Calls are categorized into InteractiveHigh for instant UI responses or AnalysisBackground for indexing. File mutations are strictly serialized to 1 write worker.

3. Neural Cache Lookup

Simplicio checks local SQLite FTS and embedding memory before inspecting disk files. Architectural decisions and previous context are retrieved in nanoseconds.

4. Bounded Context Slicing

Instead of dumping 5,000-line files, Simplicio calculates the exact token budget and yields a structural map with only relevant symbols.

5. Mechanical Edit & Hash Verification

Target blocks are replaced atomically with zero output token regeneration. SHA-256 pre-image and post-image hashes guarantee zero regression.

6. Delivery Barrier & Receipt Verification

Automated local checks run (linters, test suites). A verifiable execution receipt is returned to the agent confirming completion.

CPUExecutor

CPUExecutor & Concurrency Governance

Simplicio guarantees stability and eliminates race conditions through strict internal multi-lane governance:

Lane	Priority	Concurrency	Operations
InteractiveHigh	High	CPU × 2	Your AI agent (Claude, Cursor, Antigravity, etc.) sends tool invocations via standard JSON-RPC 2.0 without leaving the chat.
AnalysisBackground	Standard	Tokio Semaphore	Simplicio checks local SQLite FTS and embedding memory before inspecting disk files. Architectural decisions and previous context are retrieved in nanoseconds.
write_workers: 1	Serialized	1	Target blocks are replaced atomically with zero output token regeneration. SHA-256 pre-image and post-image hashes guarantee zero regression.

Mechanical Edits

Deterministic Mechanical Edit Engine

Standard LLM coding assistants rewrite entire files, burning thousands of output tokens and risking hallucinations. Simplicio enforces atomic mechanical edits:

Full-file generation

Standard LLM coding assistants rewrite entire files, burning thousands of output tokens and risking hallucinations. Simplicio enforces atomic mechanical edits:

Targeted mechanical edits

Target blocks are replaced atomically with zero output token regeneration. SHA-256 pre-image and post-image hashes guarantee zero regression.

// Simplicio
{
  "TargetFile": "/path/to/src/auth.rs",
  "TargetContent": "fn verify_token(token: &str) -> bool {\n    true\n}",
  "ReplacementContent": "fn verify_token(token: &str) -> bool {\n    validate_jwt_signature(token)\n}"
}

Agent Setup (MCP)

Agent Integration (MCP JSON)

Configure Simplicio in your favorite AI agent in less than 60 seconds:

Verify in your application

This page cannot inspect the MCP connection in your application.

Manage account

Try asking

Map the project architecture.

Apply a targeted edit and verify the result.

Run the project tests and show any failures.

FAQ & Privacy

Frequently Asked Questions & Boundaries

Is my source code sent to Simplicio servers?

No. Simplicio runs 100% locally on your machine as an MCP server. Your source code and edits never leave your machine unless your chosen model provider sends them over your configured LLM API.

What models are supported?

Simplicio is model-agnostic. It works seamlessly with Claude 3.7/Sonnet, GPT-4o, DeepSeek V3/V4, Gemini 2.0 Pro/Flash, Qwen 2.5, and local Ollama models.

How does Simplicio achieve up to 96% token savings?

By replacing repetitive full-file reads with bounded structural maps and eliminating whole-file generation in favor of targeted mechanical string replacements verified by SHA-256 hashes.