Source: https://harnsy.dev/whats-new > What changed in each harnsy release: new features, improvements and what to do when you upgrade. # What’s new What changed in each harnsy release, newest first. [v0.10.0](#v0-10-0)October 19, 2026Latest ## A guided start, and your lead on every page harnsy 0.10.0 makes your first minutes a guided start, keeps your lead one key away on every page, and lets you chat with a live agent right on the dashboard. Your AI team gets easier to run: hire from a signed catalog, connect outside tools, plan with a map of hypotheses, see where your tokens go, and work in a light theme. Highlights - A guided first run. Pick your language, sign in to Claude Code, name your company and choose your project folder. A lead studies the project first, proposes a plan and a first goal, and builds a small team in a chat, one question at a time. Name your first project in the install request and the lead starts from your words. After the first finished task the lead writes “Your first result”: what you asked, what was done, where to look. “Skip — I know harnsy” and “Show everything” are always there. - Your lead on every page. A cleaner menu keeps Home, Tasks, Teams, Agents, Organizations and Settings at hand. The lead's chat opens from any page with `Ctrl+\` (`⌘\` on Mac); `Ctrl+Shift+Space` dictates and `Ctrl+Enter` sends. A project without a lead offers “Start the lead”. - Talk to a live agent on the dashboard. For Claude Code and Codex agents the chat shows every tool call, command and edit as it happens. Your comments and decisions on task cards appear in the chat as your own messages. File paths in the chat open in a viewer when you click them, and “\[Image #N\]” marks show the pictures themselves. Pictures you send are shrunk for the agent, so they do not eat its memory; you keep the original. - Never miss what needs you. Home “Waiting for you” is cleaner and adapts to any window width. When an agent asks you to run a command, it is a request on Home and in the chat: confirm the exact command and harnsy runs it in that agent's Claude Code window and returns the output. When Claude Code itself asks you to confirm something risky, you answer Yes or No in harnsy. The lead and the Assistant do what they can themselves; what only you can do arrives as one request with a button. - Hire from a signed catalog. Employee profiles with avatar, character and skills shown before you hire. A lead can hire the role the work needs, from the catalog only. “Check for updates” finds newer versions of the people you hired; an update shows what you edited and keeps your changes unless you take the new ones. - Connect outside tools. External MCP servers by key or OAuth sign-in, a built-in catalog of 28 services, tool review, access per project and role, and a call log. Agents call the tools you ticked; deletes and secrets always wait for you. - Plan and decide. A map of hypotheses in every project: cycle goals, hypotheses to check, defer or reject, and a weekly review the lead prepares for you. Agents can offer 2–5 options with pictures, and you pick one. Task statistics, `#tags` on tasks, and per-team handover thresholds. - See where tokens go. “What eats the tokens” shows, per role, what wastes context, with advice. “Review the spend” asks the lead and the team for a plan, and you accept the changes. - Teams and subscriptions. Rename, merge, put to sleep and wake teams. Change a project's subscription and move its running agents. An agent that hit a subscription limit says until when and offers to switch. Personal and corporate Claude subscriptions stay separate, with “Not the same” to keep them apart. - Look at what agents make. Attachments open the way they should: tables for CSV and JSON Lines, a collapsible tree for JSON, colored diffs, highlighted code, Mermaid diagrams, SVG, Jupyter notebooks, Excel sheets, video and audio with a player. - Without a key. harnsy runs two projects in one organization, with no limit on agents. - And more. A light theme. Settings in sections with search. A node button in the top bar that shows outside access, full control and the subscriptions you use most. Heavy builds wait for free memory, with a “Lower to N GB” button if they wait long. “History in git” makes the lead commit each accepted task. Role memory, OpenCode 2.x, `harnsy update`, Prometheus metrics, a tray icon on Windows, and sharing your company to the gallery at harnsy.dev with a harnsy Max key. **Upgrading.** Run the install command again: your data migrates itself. A new install starts agents with permission checks until you answer the shield in the top bar. If you run several machines together, update every machine, the master first. Going back to an earlier version? Read the release page first. [How to update →](https://harnsy.dev/getting-started#update) Full release notes on GitHub ↗ ## Earlier releases - [v0.9.0](https://harnsy.dev/whats-new/0.9.0)October 7, 2026Your agents in one console[v0.9.1](https://harnsy.dev/whats-new/0.9.1) - [v0.8.0](https://harnsy.dev/whats-new/0.8.0)September 30, 2026Nothing waits for you in silence - [v0.7.0](https://harnsy.dev/whats-new/0.7.0)September 29, 2026Dictate, pause a team, choose how each role starts - [v0.6.0](https://harnsy.dev/whats-new/0.6.0)September 29, 2026Answer your agents from the dashboard - [v0.5.0](https://harnsy.dev/whats-new/0.5.0)September 27, 2026A team that runs without you dispatching it - [v0.4.0](https://harnsy.dev/whats-new/0.4.0)September 27, 2026Know what needs you - [v0.3.0](https://harnsy.dev/whats-new/0.3.0)September 26, 2026The first public build ## Your first team is one prompt away. Installing takes one sentence to your agent. To run your agents, harnsy drives a terminal: tmux if it is installed, otherwise its own terminal host (on Windows, always its own). `Install harnsy following https://harnsy.dev/llms.txt` harnsy: 1 organization, 2 projects, no limit on agents, one machine.