Truesight — Documentação de Plataforma de Avaliação de IA
Goodeye-Labs/truesight-docs
Documentação completa para avaliações de IA com critérios customizados e endpoints ao vivo
Ver no GitHubDocumentação oficial do Truesight, uma plataforma de avaliação de IA que permite definir critérios customizados para julgar saídas de modelos de IA. Oferece recursos para upload de datasets, análise de erros, construção de avaliações sem código e integração via MCP com Claude, Cursor e outras ferramentas.
Instalação
npx convez-ai Goodeye-Labs/truesight-docs1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Avaliar qualidade de outputs de modelos de IA contra critérios específicos do domínio
- Integrar avaliações de IA em pipelines de desenvolvimento via API REST ou MCP
- Implementar análise de erros e descoberta automática de critérios em datasets
Veredito

Documentação técnica bem estruturada de uma plataforma madura de avaliação de IA. Serve principalmente como referência para quem já usa ou integra Truesight; tem valor limitado como ferramenta standalone sem acesso à plataforma subjacente.
README
A tradução para o português deste README ainda está na fila.
Truesight Documentation
Official documentation for Truesight -- an AI evaluation platform that helps you build expert-grounded evaluations for your AI applications.
What is Truesight?
Truesight lets you define what "good" and "bad" look like for your specific use case, then deploy those criteria as live API endpoints. Instead of relying on generic metrics, you score AI outputs against your own domain expertise -- in real time, at scale.
Key capabilities:
- Upload datasets of AI inputs and outputs (CSV, JSON, JSONL)
- Run Error Analysis to discover evaluation criteria from your data
- Build pass/fail evaluations using Guided Setup -- no coding required
- Deploy evaluations as live REST API endpoints
- Integrate via MCP with Claude, Cursor, VS Code Copilot, and Windsurf
- Review results and promote corrections back into your datasets
Documentation
| Page | Description |
|---|---|
| Overview | Core concepts and workflow |
| Getting Started | Set up your account and run your first evaluation |
| Datasets | Upload and manage evaluation data |
| Error Analysis | Discover evaluation criteria from your data |
| Evaluations | Build and configure evaluation criteria |
| Results & Review | View results and manage the review queue |
| Teams & Organizations | Manage members, API keys, and credits |
| API Integration | Score AI outputs via REST API |
| Platform API Keys | Create and manage platform API keys |
| MCP Integration | Connect AI assistants via Model Context Protocol |
Quick links
- Product: truesight.goodeyelabs.com
- API base URL:
https://api.truesight.goodeyelabs.com - MCP server URL:
https://api.truesight.goodeyelabs.com/mcp/ - Agent skills: Goodeye-Labs/truesight-mcp-skills
Using with AI assistants
This repository is indexed by Context7, making Truesight documentation available to AI coding assistants via the Context7 MCP.
For direct MCP integration with Claude, Cursor, VS Code, and Windsurf, see the MCP Integration guide.
License
The documentation in this repository is proprietary to Goodeye Labs and provided under the Truesight Documentation License. You may read and use it to work with Truesight, allow AI and documentation tools such as Context7 to index it, and freely reuse the code samples. Redistributing or creating derivative documentation is not permitted. See LICENSE for details.