What's changed in NodeStorm.
Workflows can now look at images. The new Image Analysis node evaluates up to 10 images against your instruction with a vision model. Point it at URLs, files from your storage, or images a parent node found. The scraper node can also keep links and image sources from pages it scrapes, so scrape-then-analyze chains work end to end.
The scraper retries protected sites through a premium proxy automatically, and remembers which domains need it. LLM provider hiccups now retry with backoff instead of failing the node. And when one item in an iteration fails, the rest still deliver instead of stalling the whole run.
You can now use NodeStorm without a credit card. Free accounts get 100 node executions a month, 2 emails a day, a weekly schedule, and 10MB of file storage. Access tokens for the API and MCP are open to every tier too.
nodestorm-mcp is published on npm. Add it to Claude (or any MCP client) and you can list, build, and run your flows from a chat. The Access Tokens page shows exactly where the config snippet goes.
Trigger any flow with a POST request and your access token, optionally overriding the entry node's input for that run. You get a run id back to poll for results. The flow editor's new API dialog has a ready-to-copy example.
The homepage, dashboard, flow editor, chat, and run pages all moved to our new Glass design system. One consistent visual language and a real dark mode, instead of three UI kits stitched together.
The chat agent now lives on every flow edit page. Ask it to add nodes, rewire connections, or rewrite a step. The canvas updates in real time as it works.
New tiers: Lite $9, Starter $19, Pro $49. Same caps you've been used to, just simpler tiers. Existing subscribers stay on their current plan.
Chats are now grouped into conversations with a sidebar. You can come back to an old thread, rename it, or pick up where you left off. Full-bleed layout with collapsible sidebars on the chat page.
When a workflow fans out to multiple LLM calls in parallel, NodeStorm now staggers them. Fewer 429s from OpenAI, fewer mysterious failures, and if one upstream node does fail, downstream siblings skip cleanly instead of erroring.
The Email node now produces raw HTML directly via a two-call architecture. Output looks better, renders consistently across email clients, and uses gpt-5.4-mini for faster generation.
You can now feed a URL into an HTTP request node from the previous step's output. Useful for "fetch this page, then fetch every link on it" style flows.
The "Generate a workflow from a description" feature now references a library of known-good flows when building yours. Better suggestions, fewer bizarre node combos.
Billing portal access, branch routing on conditional nodes, usage stats display, edge arrow rendering, and form autofill. Plus chat resilience improvements: truncation detection, partial-turn persistence, and a working indicator.
Send and receive messages through Telegram from your workflows. Channel-agnostic adapter with deep-link verification. Works whether or not your users have Telegram Desktop.
