Selectools

Selectools is a Python framework designed for building production-ready AI agents entirely in plain Python. It addresses common challenges faced by agent development teams, such as debugging, data privacy, and cost tracking, without requiring proprietary SaaS accounts, custom DSLs, or external dependencies. The framework is open source, with a stable public API, and is delivered as a single pip install package maintained by NichevLabs.

Key features of Selectools include support for multi-agent graphs, tool calling, retrieval-augmented generation (RAG), and a suite of 50 evaluators. It offers a visual drag-and-drop builder, one-line REST API serving, and built-in audit logging. Guardrails are integrated for PII redaction, topic blocking, toxicity detection, and prompt injection screening, including 15 detection patterns and LLM-based coherence checks. Every agent run returns a structured trace and reasoning, providing a detailed timeline of LLM calls, tool selections, and executions for transparent debugging.

Selectools integrates with major AI model providers such as OpenAI, Azure OpenAI, Anthropic, Gemini, and Ollama, and supports over 100 models via LiteLLM. The framework includes automatic usage tracking with per-call token counts and cost calculations across 115 models, leveraging built-in pricing data. It also offers features for provider failover, streaming, budget control, and supervision, as well as support for customer support and research team workflows.

The platform is aimed at developers and teams seeking a production-grade agent framework that remains within the Python ecosystem and does not require learning new languages or managing multiple packages. Selectools is open source and does not require payment for SaaS or debugging tools. It can be deployed locally or served via a one-line command, and its guardrails and audit features are configured directly in the agent constructor for ease of use.