Skip to content

OpenCognit 0.6.0 — Atlas

Latest

Choose a tag to compare

@Panto92 Panto92 released this 19 Nov 00:31
· 1 commit to main since this release
dbdc6ef

Atlas — OpenCognit 0.6.0

OpenCognit 0.6.0 — Atlas

Your project's brain at a glance: the graph now shows your project as an Atlas — a brain map where every area of the code is a lobe —, as Layers from agents through areas to outcome, and the Neurons are sorted instead of tangled. Plus one field for tasks and commands, a calmer browser and lots of polish.

New: Atlas

  • Your project as a brain map across the whole view: every area a lobe of cells, sized by its file count
  • Lobes light up where agents work; every agent stands at its lobe with a name tag ("Claude · editing main.ts", "waiting for you")
  • Two agents in the same area: red ring
  • Hover shows files, changes today and who is there; click highlights, click an agent to open its chat

New: Layers

  • Built like a neural network: agents on the left, project areas in the middle, outcome on the right — to review, tests, commits today, needs you
  • Pulses run left to right while agents work; anyone waiting for you has an amber thread to "Needs you"

Neurons, sorted

  • Neurons are the graph's start view; the Cosmos view is gone (the timeline still replays it)
  • Related areas (imports, changed together) sit side by side, each in its own colour
  • One stem per subfolder, every tip a file — large folders spread over several stems; subfolder names appear as you fly closer
  • Links only where there is a real relationship
  • Click an agent's name tag to open its chat

One field for tasks and commands

  • "Task or command …" at the top — click, Ctrl+K or Ctrl+Shift+A open the same box
  • A sentence becomes a task: to the suggested agent (with the reason, own worktree, review card), as variants to two agents, or to the orchestrator
  • A short command word shows commands first

Browser in the side panel

  • The address bar shows just the domain with a lock; click for the full address
  • Reload next to back/forward; auto-reload, design overlay, developer tools and "Open in system browser" behind "…"

Many agents, smoother

  • Terminal output is batched and unfocused terminals draw less often — noticeably calmer with 32 agents
  • If the interface crashes, it reloads itself; your agents keep running
  • Kimi and Hermes report "waiting" again after every answer; Gemini no longer hangs at startup on the browser MCP

Polish

  • One type scale, slightly larger; buttons with a press feel, surfaces instead of lines, calmer task cards
  • "While you were away" with a headline and timeline; notifications without duplicates
  • Agents tab as a work stream: who needs you first, then finished work, idle ones folded
  • Changes: new files are no longer pre-selected and are folded by folder — no accidental commit of thousands of files
  • Model picker no longer jumps; reasoning effort applies to the selected model
  • Side panel tab order survives a restart
  • Tour in 4 steps instead of 16
  • Smooth splash screen