// the take
LLMStack is a self-hostable, no-code builder (Django backend, React frontend) for chaining LLM calls into agents, chatbots, and RPA-style workflows, with built-in data ingestion and vectorization. It's aimed at teams who want to wire up multi-step LLM apps without writing Python, and the company behind it also sells a hosted version (Promptly).
Data ingestion handles multiple formats (CSV, PDF, DOCX, PPTX) from multiple sources (Google Drive, Notion, direct upload, web) and does chunking/embedding automatically instead of leaving that to the user. Apps built in the no-code editor are exposed over an HTTP API and can be wired into Slack or Discord, so they aren't locked into a single chat widget. Provider support spans OpenAI, Anthropic, Cohere, Mistral, Stability, and local models, so a workflow isn't tied to one vendor's API once built.
Last commit was December 2024 — nearly two years with no activity as of late 2026, which for a project whose vendor now pushes a hosted SaaS (Promptly) is a strong signal the open-source build is no longer the priority and may be effectively abandoned. It ships with a default admin account (admin/promptly) that the README just says to change after login, which is exactly the kind of default credential that gets left in place on self-hosted instances. Running it for real means standing up Django, Postgres, a vector store, and a Docker-dependent job runner — more infrastructure than a 'pip install and go' pitch suggests. No-code chaining tools like this hit a ceiling fast once a workflow needs custom branching logic or non-trivial error handling that the visual builder doesn't model, forcing you into writing custom processors anyway.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →