The developer of PromptChain, a Streamlit app that chains a small local model into a larger coding model, has released several updates based on user feedback. The tool automatically unloads VRAM between models to allow operation on 8–16 GB cards without manual swapping.
- Reasoning models now properly handle <code><think></code> blocks and DeepSeek-R1 / Qwen3 reasoning deltas in a separate collapsed panel.
- Multi-file output renders as per-file tabs with zip download or save-all-to-folder options.
- Pipeline profiles allow saving and switching entire setups including backends, models, temperatures, and system prompts.
- Persistent single-model chats enable ChatGPT-style pages for the Prompter or Coder independently.
- Quick mode skips the review step to go directly from idea to code.
- Refine-in-place allows follow-up instructions to edit existing code with diffed version history.
The tool supports local-first workflows with no telemetry and works with LM Studio, Ollama, or any OpenAI-compatible server.