Promptchain: Agents Manage Local Models Directly
atharva557 · reddit · 2026-07-19
The author has spun out the "model switching core" from their Streamlit app into a standalone Python library and MCP Server: promptchain (MIT, alpha v0.1).
It primarily solves the issue where backends like LM Studio, Ollama, and llama-swap manage their own model residency and unloading individually, but lack a unified client for explicit VRAM control across multiple backends. This library provides a unified stream() interface and model manager capable of evicting models based on priority/LRU/residency limits, while also supporting pinning, preloading, and atomic unloading/loading.
The author also highlights the MCP Server's utility: once connected to Hermes Agent, Claude Code, or other MCP clients, agents can manage GPUs directly via natural language (e.g., "unload drafter, load qwen3-coder"). The project is open-source, with a repo, examples, and PyPI installation provided.
More from coding & agent
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22