January 7, 2026
Read-only or read-the-room?
Show HN: Open-source AI workflows with read-only auth scopes
DIY AI workflows go 'read-only'—fans cheer safety, critics say 'why bother'
TLDR: Seer, an open-source AI workflow builder, pushes read-only permissions to keep your accounts safe. The crowd split: security fans love the restraint, while automation diehards say it’s useless without write access, with extra debate over heavy setup versus easy one-click deploy and the split between agents and workflows.
Seer just landed on Show HN with an open‑source builder for automated AI workflows—and the “read‑only permissions” pitch set the comments on fire. Supporters shouted “finally!” at an app that won’t demand full control of your Gmail, GitHub, or Google Drive, while skeptics snarked that an automation tool that can’t write is like a robot butler who won’t touch the dishes. The team’s big design stance—keeping workflows (step‑by‑step routines) separate from agents (chatty assistants)—got nerd applause for clarity, and eye‑rolls from folks who think everything should be one giant bot. The setup story drew memes: one‑command “uv run seer dev” was hailed as “press play for devs,” but others groaned at spinning up Docker, Postgres, Redis, and a worker just to try it. The one‑click deploy to Railway with a $15–30/month estimate sparked budget banter and “one‑click to surprise bill” jokes. Folks liked the visual editor, Markdown exports, and clean FastAPI backend; some roasted the name clash (PyPI “seeragents” vs CLI “seer”) with “Seer? I hardly know her.” Privacy hawks loved the read‑only auth scopes framing, while power users demanded toggles to unlock writes when needed. And yes, requiring an OpenAI or Anthropic key reignited the classic “open‑source, closed wallet” debate. Drama level: medium‑spicy, with extra salt on permissions.
Key Points
- •Seer is an open-source workflow builder with visual editing and AI-assisted development for automated workflows.
- •The project separates workflows (deterministic, node-based) and agents (dynamic, message-based) with distinct APIs and components.
- •Quick start uses "uv run seer dev" to spin up Docker services (FastAPI backend, PostgreSQL, Redis) and open the builder in a browser.
- •One-click deployment to Railway sets up a FastAPI backend, background worker, PostgreSQL, and Redis; setup requires an OPENAI_API_KEY and costs ~$15–$30/month.
- •Configuration uses a .env file with required LLM keys (OpenAI or Anthropic) and optional integrations (Google Workspace, GitHub, Supabase, Tavily); development uses a Taskiq worker and Uvicorn hot-reload.