Skip to main content

Version 3.1.0

Release date: October 14th, 2026

Version 3.1.0 brings a redesigned interface, a guided Discovery setup, Branches in place of changesets, improved Skills, MCP connections and Artemis Desktop for running models on your own machine. Optimise has been retired.

A redesigned interface​

  • A new layout with a collapsible sidebar, back and forward navigation, and your account at the foot of the sidebar.
  • A theme switcher, with dark mode as the default.
  • Search or run an action (⌘K): find Plans, Branches and Discovery Runs, jump to recent items and run page actions.
  • All settings are in one sidebar. Project settings have their own section, and the project header has a shortcut to them.
  • Files open as tabs, with a preview tab, and images are previewed.
  • Redesigned list pages with an expandable search.
  • Markdown in chats and pages renders diagrams, images and videos.

Discovery: guided setup and next steps​

  • Starting a Discovery Run is now a four-step setup: Evaluation, Objective, Preferences and Summary. The objective, preferences, Version budget and approval mode are all set there.
  • Metric importance is set in tiers, and each Metric's direction (Higher is better or Lower is better) is shown and can be edited during setup. Changing a Metric's direction rescores the runs that use it.
  • AI-assessed Metrics are optional and off by default for new runs.
  • Reuse a measured Baseline from an earlier run, or measure a fresh one during setup.
  • When a run finishes, it suggests where to go next, and each suggestion can be started directly.
  • Score any Version on request with the run's own judge, including Versions you added yourself. When a Version has no score, the reason is shown.
  • Split a Version into smaller, separate changes from its actions menu.
  • Runs record why they failed, and Versions explain their approach, change and conclusion.
  • Cancel a run from the run list, search runs by objective, and see a Title column in the Versions table.
  • Experiments have a short ID such as EXP-3, and you can reference Experiments and Versions with # in chat.
  • The canvas has been rebuilt, with an experiment queue in the toolbar. Experiment views are in tabs, the Experiments list can be sorted by column, and the board supports drag and drop.
  • Experiments have a results section with their Metrics and a forest plot.
  • Download a Version as a patch, and pick a model directly when combining or generating Versions.
  • Attach files to the prompt when creating a run.
  • New sample projects: Particle Life, Julia Set, Smoke and Shop Analytics (SQL).
  • Up to 500 benchmark repetitions per run.

Branches​

Changesets are now Branches, with a redesigned page for each Branch.

  • A Metrics tab replaces Compare. It shows each Metric separately rather than one averaged verdict, with grouping, filters and sorting.
  • Take a Branch from its first run to a measured result, or start a Discovery Run from a Branch and reuse its measurements.
  • In the Changes tab, collapse all files and mark files as viewed. Change counts update live while an agent writes code.
  • The chat working on a Branch opens from the Branch's toolbar.
  • Find the Branch behind a pull request, Git branch or commit, and search Branches by name.
  • Pull request descriptions are written from the change's recorded results.
  • Administrators can publish a Branch created by another user.

Skills​

  • Settings → Skills is redesigned around an AGENTS.md file and your Skills: instructions your agents read before you ask them anything, and focused notes they open only when needed.
  • Skills appear in the chat composer, and system skills can be run as slash commands. /compact is now /system:compact.
  • Tailor to this Project adapts a Maintain Rule to your project.
  • Agents now get context about your repository automatically: its languages, the kind of project, and any files written for AI tools.
  • The Artemis Skills for your own coding agent include a new quickstart skill for new users, new repositories and existing projects. Connect Your Agent uses it.

MCP connections​

  • Connect your agents to external tools under Settings → MCP Connections, including sign-in with OAuth. The catalogue includes integrations such as Granola and GitLab.
  • Add GitHub or Bitbucket MCP servers from a saved Git key, and test the key before adding it.
  • Each server lists its tools in a searchable table, and you can search tools across all servers.
  • Copy your connections as JSON (mcp.json) and import them elsewhere.
  • MCP connections are available from the + menu in the chat composer.
  • See MCP Connections.

Artemis Desktop​

Run models on your own machine and use them in Artemis.

  • The Artemis Desktop app connects models running on your computer to the platform. Installers are available for Windows and Debian.
  • Your local models appear in the model picker. A model shown in green in the desktop app is ready to use in Artemis.
  • If your desktop app is too old for the platform, it tells you to update.
  • The Artemis CLI can use your local models.
  • See Artemis Desktop.

Maintain​

  • Issue lanes are saved, every move is recorded, and Issues can be dragged between columns.
  • Fix an Issue with a coding agent or in a Discovery Run. The Fix dialog recommends one, checks that your validation Script covers the finding, and asks for benchmark repeats.
  • Fix Issues from your own checkout with the CLI. One local fix can resolve several Issues.
  • Only one fix runs at a time per Issue, and a running fix can be cancelled.
  • Archive and restore Issues from an Archive page, with a list view and actions for each Issue.
  • Rule Library: save Rules to your own library and add them to any project.
  • Rules are written in Markdown. Create rule from markdown is now Create custom rule, and deleting a Rule warns that its Issues are deleted too.
  • Scan summaries show which findings are already on the board or archived, and View findings filters the board by Scan.
  • Filter Issues by Branch or Discovery Run. Board filters are kept per project.
  • A Files view on the Issue board, and one chat shared by the Issues and Rules pages.
  • Every fix attempt is kept on its Issue, and done Issues can be fixed again.
  • Scans can run without Rules, and the Maintain agent can find Issues by Rule, raise Issues from chat, and run Scans and fixes itself.

Plan​

  • The Build tab is now Plan, with the Plan's status and Git status in the toolbar.
  • Rename Plans from their menu, and search Plans by name and prompt.
  • Building a Plan runs the project's checks, and the Plan's chat can read the results.
  • Plans connect to Discovery, and the planner offers a Build Plan action.
  • Attach files when creating a Plan or a Task.

Chat and agents​

  • Agents has its own page in each project and across all your projects.
  • Slash commands, with a picker in the composer.
  • Composer modes for Maintain, Discover and a new Code mode.
  • Paste or drop images and files onto the prompt.
  • Suggested prompts based on what you are doing, and links to tutorial videos for introductory questions.
  • The agent can suggest Discovery settings, which you can apply from the chat.
  • Fork a conversation. Search, sort and filter chats, with live status.
  • Conversations recover after a dropped connection.
  • The cost and context used by a conversation are shown in the prompt footer.
  • Agents handle slow models and dropped responses more reliably.

Validations and Metrics​

  • Validation Scripts run in phases, with repeats that adapt to the results and a report for each pass. A Script needs at least one command.
  • Scripts always run on the project's Runner.
  • The Metrics table recommends how many more repeats would settle a comparison, shows each Metric's definition on hover, and lists recorded values for every Metric.
  • The wording is consistent everywhere: Higher is better or Lower is better, Baseline, and composite score.
  • A Script without a benchmark command runs once.

Project overview and settings​

  • The project overview is the project's landing page, with counters, what the agents are doing, and Getting Started cards that tick off as you complete them.
  • Item counts in the sidebars.
  • Project setup offers a local agent option, and agent and Runner setup is simpler.
  • A New menu on the Projects page, including uploading a .zip file.

Models​

  • Artemis Desktop models in the model picker.
  • Edit a model's base model ID and default reasoning effort.
  • A profile can override its prompt-cache duration.

Accounts and sign-up​

  • Registration and password reset are protected by a captcha.
  • Email verification links now expire.
  • Invited users receive a link to set their password, and administrators can share that link when creating a user.

Artemis CLI and Runner​

Artemis CLI 1.1.18

  • Maintain commands for agents and CI, local fixes, archiving Issues and cancelling a running fix.
  • Drive the Discovery setup, reuse a measured Baseline and score a Version with the run's judge.
  • New compare commands: changeset compare and discovery compare.
  • Artemis Desktop models are supported.
  • Requests are retried through brief network interruptions, and destructive commands need confirmation.
  • Signed Windows ARM64 builds and a PowerShell installer.
  • New --ssl-cert-file and --ssl-verify options.
  • The context commands are renamed skills, and the task commands are retired.
  • JSON output has changed: keys are camelCase, errors are a single JSON object, and results include web links. Scripts that parse CLI JSON may need updating.

See Artemis CLI.

Artemis Runner 5.3

  • --non-interactive for agents and scripts.
  • Upgrades work without a reachable release server.
  • Runner names longer than 59 characters are rejected.
  • The Runner now runs Branch validations only.

See Artemis Runner and Downloads.

API​

  • The Artemis API is now served under /api/v1. Update any custom integrations that call the API directly.
  • The Python client (artemis-client 2.2.3) can retry when Artemis is unreachable.

On-premise deployments​

  • Follow the upgrade instructions provided with the release.
  • New backup and restore commands give a safe path through upgrades.
  • Sizing guidance and a pre-deployment check.
  • Optional CPU autoscaling on Kubernetes.
  • Security hardening across the platform.

Discontinued​

  • Optimise has been removed, including the Optimise tab.
  • The CLI task commands are retired.
  • The local Windows installation wizard is removed.
Last updated on by Neda Zarei