Pular para o conteúdo
ReviewRamble
ReviewRamble

A newsletter semanal

IA, analisada, na sua caixa de entrada.

  • Novas análises de ferramentas, modelos, servidores MCP e repositórios, com o veredicto logo no início
  • Preços e notas reconferidos quando algo muda
  • Novos artigos de análise assim que são publicados

Modelos, agentes, servidores MCP e as ferramentas que realmente valem seu tempo.

Um e-mail por semana. Cancele pelo link em qualquer edição. Como usamos seu e-mail

A stack de IA

Dos dados brutos à IA em produção, etapa por etapa.

22 etapas, da esquerda para a direita: ingestão, análise de documentos e OCR, chunking e embeddings, bancos de dados vetoriais e grafos, agentes e roteamento, os modelos e, depois, fine-tuning, inferência, implantação, guardrails e segurança. Passe o mouse sobre uma etapa para ver as ferramentas; clique para compará-las.

Mostrar o caminho para

00Ingestion& crawling00Parsing & OCR00Chunking00Embeddings00Vectordatabases00Knowledgegraphs & memory00Retrieval& reranking00Orchestration& agents00Tools & MCP00LLM routing& gateways00Language &reasoning models00Image models00Video models00Speech &voice models00Fine-tuning00Inferenceengines00Deployment& hosting00Guardrails00Security &red-teaming00Evaluation00Observability& tracing00Apps &interfaces

Role para o lado para acompanhar todo o pipeline →

Etapa 00 · Preparar

Ingestão e crawling

Puxar conteúdo de sites, arquivos, apps e bancos de dados. Um modelo só é tão atual quanto os dados que você fornece a ele. A ingestão decide o que o seu sistema sabe e quão atualizado ele se mantém.

Comparar as 12 ferramentas

Ferramentas que avaliamos

  • Crawling de sites públicos: um crawler que devolve Markdown limpo economiza uma etapa de parsing.
  • Apps e drives internos: prefira conectores com sincronização incremental a reimportações completas.
  • Respeite o robots.txt e os termos do site; rastreie apenas o que você tem permissão para rastrear.

Casos de uso, etapa por etapa

As etapas de que cada tarefa precisa, em ordem, com as ferramentas por onde começaríamos.

Chat with your documents

Answer questions over PDFs, wikis and policies, with sources.

  1. 1Ingestion & crawlingFirecrawl4.2
  2. 2Parsing & OCRDocling4.5, LlamaParse4.2
  3. 3ChunkingChonkie3.7
  4. 4EmbeddingsQwen3-Embedding and Qwen3-Reranker4.2
  5. 5Vector databasesQdrant4.7
  6. 6Retrieval & rerankingVoyage rerank-34.6
  7. 7Orchestration & agentsLlamaIndex4.3
  8. 8Language & reasoning modelsClaude Sonnet 5.54.6
  9. 9Inference enginesTogether AI4.5
  10. 10Deployment & hostingModal4.5
  11. 11GuardrailsNVIDIA NeMo Guardrails4.0
  12. 12Security & red-teamingpromptfoo4.5
  13. 13EvaluationRagas3.4
  14. 14Observability & tracingLangfuse4.6
  15. 15Apps & interfacesOpen WebUI4.6

Customer support agent

Triage tickets, answer from your help centre and hand off to people.

  1. 1Ingestion & crawlingApify4.1
  2. 2Parsing & OCRMarkItDown3.8
  3. 3ChunkingChonkie3.7
  4. 4EmbeddingsCohere Embed 54.2
  5. 5Vector databasespgvector4.6
  6. 6Retrieval & rerankingCohere Rerank 43.7
  7. 7Orchestration & agentsLangGraph4.6
  8. 8LLM routing & gatewaysJev3.8
  9. 9Language & reasoning modelsClaude Haiku 5.54.5
  10. 10Deployment & hostingAWS MCP Server4.4
  11. 11GuardrailsNVIDIA NeMo Guardrails4.0
  12. 12Security & red-teamingpromptfoo4.5
  13. 13EvaluationDeepEval4.2
  14. 14Observability & tracingLangSmith4.2
  15. 15Apps & interfacesIntercom Fin4.7, Sierra4.5

Coding agent

Plan, write, test and review code across a repository.

  1. 1Ingestion & crawlingGitHub MCP Server4.7
  2. 2Parsing & OCRMarkItDown3.8
  3. 3ChunkingLlamaIndex4.3
  4. 4EmbeddingsQwen3-Embedding and Qwen3-Reranker4.2
  5. 5Vector databasesLanceDB4.0
  6. 6Tools & MCPGitHub MCP Server4.7, Context74.5, Serena4.2, CodeGraph4.1
  7. 7LLM routing & gatewaysOpenRouter4.6, LiteLLM4.3
  8. 8Language & reasoning modelsClaude Opus 5.54.7, GLM-5.34.4
  9. 9GuardrailsJev MCP4.3
  10. 10Security & red-teamingSnyk MCP Server (Snyk Studio)4.0, promptfoo4.5
  11. 11EvaluationInspect4.4
  12. 12Observability & tracingLangfuse4.6
  13. 13Apps & interfacesClaude Code4.7, Cursor4.5, OpenCode4.6

Research & knowledge graph

Connect people, projects and events across thousands of documents.

  1. 1Ingestion & crawlingCrawl4AI4.2
  2. 2Parsing & OCRDocling4.5, PaddleOCR-VL-1.64.4
  3. 3ChunkingChonkie3.7
  4. 4Embeddingsvoyage-4-large4.5
  5. 5Vector databasesWeaviate4.2
  6. 6Knowledge graphs & memoryGraphRAG3.5, Neo4j4.1
  7. 7Retrieval & rerankingJina Reranker v3.54.3
  8. 8Orchestration & agentsLlamaIndex4.3
  9. 9Language & reasoning modelsGemini 4 Argon4.5
  10. 10EvaluationRagas3.4
  11. 11Observability & tracingArize Phoenix4.2
  12. 12Apps & interfacesAnythingLLM4.3

Assistant with long-term memory

Remember users, preferences and facts across conversations.

  1. 1Ingestion & crawlingUnstructured3.8
  2. 2Parsing & OCRMarkItDown3.8
  3. 3ChunkingChonkie3.7
  4. 4EmbeddingsJina Embeddings v54.3
  5. 5Vector databasesQdrant4.7
  6. 6Knowledge graphs & memoryGraphiti4.3, Mem04.0
  7. 7Orchestration & agentsOpenAI Agents SDK4.4
  8. 8Language & reasoning modelsGPT-6.1 Sol4.6
  9. 9Deployment & hostingCloudflare MCP Server4.5
  10. 10GuardrailsNVIDIA NeMo Guardrails4.0
  11. 11Security & red-teamingpromptfoo4.5
  12. 12EvaluationDeepEval4.2
  13. 13Observability & tracingLangfuse4.6
  14. 14Apps & interfacesLibreChat4.4

Voice agent

Answer calls and talk in real time.

  1. 1Ingestion & crawlingFirecrawl4.2
  2. 2Parsing & OCRDocling4.5
  3. 3ChunkingChonkie3.7
  4. 4EmbeddingsEmbeddingGemma 24.3
  5. 5Vector databasespgvector4.6
  6. 6Orchestration & agentsPydantic AI4.5
  7. 7LLM routing & gatewaysLaya3.4
  8. 8Language & reasoning modelsGemini 3.8 Flash4.1
  9. 9Speech & voice modelsParakeet TDT 0.6B v34.1, Cartesia Sonic 3.64.4
  10. 10Inference enginesGroq4.0
  11. 11Deployment & hostingModal4.5
  12. 12GuardrailsNVIDIA NeMo Guardrails4.0
  13. 13Security & red-teamingpromptfoo4.5
  14. 14EvaluationDeepEval4.2
  15. 15Observability & tracingLangfuse4.6

Document processing

Extract fields from invoices, forms and contracts, then route them.

  1. 1Ingestion & crawlingUnstructured3.8
  2. 2Parsing & OCRReducto4.3, PaddleOCR-VL-1.64.4, DeepSeek-OCR 23.7
  3. 3EmbeddingsQwen3-VL-Embedding4.3
  4. 4Orchestration & agentsn8n4.4
  5. 5LLM routing & gatewaysGLiNER2.5-Decide3.6
  6. 6Language & reasoning modelsGPT-6 Luna4.3
  7. 7Deployment & hostingAWS MCP Server4.4
  8. 8GuardrailsNVIDIA NeMo Guardrails4.0
  9. 9Security & red-teamingSnyk MCP Server (Snyk Studio)4.0
  10. 10EvaluationDeepEval4.2
  11. 11Observability & tracingOpik4.1