Changelog
What's new in CadraOS
New capabilities and improvements, newest first.
October 2026·RuntimeNew
Cadra Desktop for macOS
- One menu-bar app lets your Mac take Cadra jobs and run them with Claude Code or Codex.
- Sign in with your Cadra account. Turn job-taking on or off from the menu bar.
October 2026·AgentsNew
Harnesses and session continuity
- Give each agent a harness: the tools it gets, step limits, reply length and how it narrates its work.
- Long conversations keep going: older turns are folded into a summary instead of being dropped.
October 2026·SkillsNew
Import skills from Claude plugins
- Paste a GitHub repo with a Claude Code plugin marketplace and import its skills, files included.
- Review each skill before it is added. Refresh shows when the source has an update.
October 2026·RuntimeImproved
Remote runs that survive restarts
- Jobs on remote runners keep going when Cadra restarts, and conversations resume where they left off.
October 2026·ChannelsNew
Website chat widget
- Put an agent on your website as a chat widget.
October 2026·ConnectorsNew
Higgsfield and Google Drive
- Agents can create images and video with Higgsfield and work with files in Google Drive.
October 2026·AccessImproved
Simpler roles and invite-only access codes
- A clearer roles model for who can do what in an org.
- Sign-up can be limited to people with an access code.
September 2026·BuilderNew
Build Cadra from Claude Code and Codex
- The Cadra builder MCP lets Claude Code or Codex create and edit your agents, skills and tools.
- A new “Claude & Codex” page shows how to connect them.
September 2026·KnowledgeNew
Knowledge-base MCP and ingest
- Connect a knowledge base to any MCP client and search it from there.
- Add documents to a knowledge base with the new ingest flow.
September 2026·AINew
AI functions and the AI prompt writer
- AI functions run a fixed prompt without a full agent — for jobs like generating an image or describing a photo.
- The AI prompt writer drafts an agent's prompt for you.
September 2026·BillingNew
Platform models and subscriptions
- Use models provided by Cadra without bringing your own API key.
- Subscription plans to pay for usage.
September 2026·Agentic BoardsImproved
Boards at /work, with Jira and Notion lanes
- Boards have a new home at /work, redesigned around the cards your agents are working.
- Link a board to Jira or Notion so cards stay in step with your existing tracker.
September 2026·AppImproved
A cleaner app: Spaces, phone navigation, language
- Workspaces are now called Spaces.
- New navigation that works on a phone.
- Choose the app's language in settings.
June 2026·Agentic BoardsNew
Agent-powered boards you can drop into any app
- A new board surface where agents are the workers — drop a card, an agent picks it up, works it live, writes the result back to your system, and advances the card.
- Watch the run happen on the card itself: streaming tokens, tool calls, and artifacts, with a clear failed-run state and retry.
- Embed it in your own product with an API key and org ID — the board is owned by the OS, rendered in your app.
- Boards are mobile-ready with swipe navigation and dark mode, and update instantly with optimistic UI.
June 2026·Agentic BoardsNew
Design and ship boards without writing the engine
- A visual editor to design board lanes, transitions, and the agent, team, or webhook that handles each lane.
- Test a draft board against real agent executions before you publish it, then reset cards and iterate.
- A “get the code” recipe to host a board in any app — React snippet, config endpoints, and server setup in one place.
May 2026·ChatImproved
Chat that narrates its work as it happens
- Agents now explain each step in order — text, then the tool running, then the result — instead of a silent wait followed by a wall of text.
- Running tools show a live elapsed timer; finished tools show how long they took.
- Reloaded conversations render exactly as they streamed, with every step preserved.
May 2026·PerformanceImproved
Agents that feel instant
- Agents acknowledge work within a couple of seconds and stream continuous progress, cutting the perceived wait before the first response.
- Faster execution start-up and artifact loading across the platform.