Skip to content

Canoryn 0.2.0

Choose a tag to compare

@dpak-maurya dpak-maurya released this 27 Jul 16:06

Canoryn 0.2.0

Browser nodes that act like a real browser desk, terminals agents can actually read, models you pick on purpose, and a clearer canvas for building and running workflows.

Added

  • Browser that behaves like a browser — search a phrase from the address bar; localhost stays http for local dev servers; idle pages sleep after about 30 minutes and come back where you left them.
  • Point at a thing, hand it to an agent — click an element on a page, add a note about why, and wire it into an AI step (including multi-element picks).
  • Tell me when this changes — watch an element and let a workflow run on price drops, build status, stock arriving, and similar signals.
  • Capture a region for vision — for charts, PDFs, and anything the page structure can’t describe.
  • Terminal steps agents can read — live terminal on the canvas; an agent can run a command and see whether it passed. The shell can outlive the view.
  • Pick your models — star the few you use; those are the ones in the picker. Badges show which support tools, vision, and JSON.
  • Attach images and files in chat — several at once, including PDFs and text, through one multimodal path.
  • Clearer nodes and run insight — redesigned node chrome, execution context, per-node logs, and run history you can search with Run IDs.
  • Secrets where they belong — environment and Keychain-backed secrets for workflows that need credentials without stuffing them into the graph.
  • Richer authoring for agents — YAML source surface, canvas query language, and inspect formats (including mermaid/YAML) so CLI/MCP agents can read and edit graphs more precisely.
  • Media control on the canvas — find and silence background audio from browser sessions.
  • MCP run inspection — list runs and inspect steps from external agents without opening the UI.

Changed

  • Model selection is literal — no more provider tiers quietly choosing a different model than the one you picked.
  • Zooming no longer changes the document — wires stay attached to their ports at every zoom level; geometry stays scale-invariant.
  • Downloaded / restored workflows land disarmed — they stay off until you enable them; installs are size-capped and checksum-verified.
  • Version history is durable — more than three snapshots, and failures are reported instead of swallowed.
  • Canvas LOD and chrome — smoother detail at distance, animated active wires, and a steadier grid while you work.

Fixed

  • Workflows saved without their .cryn extension on some builds — Finder preview and double-click open work again.
  • Chat failing outright on some models when tools were attached.
  • Browser resume flicker, freeze-on-switch, and pages loading at the wrong size.
  • A page you’re looking at is never the idle-eviction victim.
  • Wire/port meeting points and low-detail node readability.
  • Terminal focus, login-shell startup, and honest store/HTTP error reporting.
  • Graph runner execution limits and cleaner LocalizedError messaging.

Install

Canoryn is still in beta and is not notarized yet, so macOS may show a one-time warning. To open it:

  • Easiest (Terminal): xattr -dr com.apple.quarantine /Applications/Canoryn.app
  • Or: System Settings → Privacy & Security → scroll down → Open Anyway

Thanks for trying the beta — support@procodeai.com.