Custodian Labs: A Python SDK to build and deploy LLM agents without wiring infra
Custodian-Labs · reddit · 2026-08-17
Custodian Labs is a Python SDK for building and deploying LLM agents, designed to eliminate the need to repeatedly set up infrastructure. Key features include:
- Model Agnostic: Switch LLM providers without rebuilding the agent.
- Built-in RAG: Connect your own files and data sources.
- Multi-agent Support: Build specialized agents that collaborate.
- Privacy/PII Layer: The Guardian Layer detects and protects sensitive data before it reaches the LLM.
- Deployment Handled: Minimizes infrastructure and configuration requirements.
Originally started as a privacy layer, the project expanded into an end-to-end SDK based on developer feedback. The author is seeking input on the biggest pain points in current agent stacks and preferences for abstraction levels.
More from coding & agent
- 5 Patterns from DeepSeek's Open-Source Agent Harness — Saboo_Shubham_ · 2026-08-17
- Top 12 open-source AI agent and LLM monitoring tools in 2026 — TheTuringPost · 2026-08-17
- Developer releases tool to remove 'AI slop' from generated text — doodlestein · 2026-08-17
- Architecture for Agents: Feature Separation & Bayesian Routing — vedp2 · 2026-08-17
- Using an Agent to Automatically Track Visa Application Status — gabriel1 · 2026-08-17
- Local AI Troll Skill Simulates Cloud Agent Failures — julianharris · 2026-08-17