Skip to main content
Download the latest version from the releases page. See Installation for updates.
​
MCP sign-ins and larger war rooms
  • MCP servers that need a sign-in get Connect. Kepler now asks a remote MCP server that does not say how it signs in whether it needs one. If it does, its row in Settings > Tools & MCP shows Sign in and a Connect button. Before, Kepler treated these servers as open, and every request failed with 401 Unauthorized. See Signing in.
  • Plugin servers that declare a sign-in work. A plugin’s MCP server that declares its sign-in the Claude Code way, with an oauth block, gets a working Connect. Installing or updating a plugin, and /plugins, say which of its servers still need a sign-in. See Servers that need a sign-in.
  • A server whose sign-in settings are wrong or unsupported shows Config error and says what to fix, instead of sending requests with no token.
  • Slack’s MCP server signs in through Kepler’s own Slack app, so the consent page names Kepler.
  • War rooms with more than four members. The room shows the lead and the first four members, and the rest keep working without a pane. Before, a fifth member took over the room: the lead, its composer and the other members disappeared. See What you see.
​
Hooks, plugins, channels and chat automations
Extend Kepler
  • Hooks. Run your own check at moments in Kepler’s work: before a tool runs, when a message arrives, when a run ends, when a watcher fires. Hooks can block or ask, never approve. Hooks written for Claude Code run unchanged. Describe one with /hook and Kepler drafts it for you to confirm. See Hooks.
  • Plugins and marketplaces. Install bundles of skills, commands, agents, hooks, MCP servers and chat channels from a marketplace, GitHub, a git URL or a folder. You see everything a plugin runs before it runs. Plugins can ask for settings, and secrets go to the keychain. Plugins in the Claude Code and Cursor formats install too. Ask with /plugins. See Plugins.
  • Customize. Settings > Customize is now one place for plugins, skills, hooks, commands and agents, with one search and a workspace filter. Commands get an editor. Agents moved here from their own settings page, and Signature moved to Settings > General. See Customize.
  • AGENTS.md. Kepler reads a workspace’s AGENTS.md alongside its kepler.md. See Memory.
Channels
  • Channel plugins. A plugin can bring a new chat channel. Add channel on the Gateway page installs one, wraps a channel server you already have, or has Kepler build one with /channels. See Add a channel.
  • Warm and cold sessions. A chat keeps its session while you keep talking. After six quiet hours the next message starts a fresh one, and /sessions and /switch take you back. Each thread is its own session. See Which session a message goes to.
  • Kepler can post first. In Assist and Yolo, Kepler can send a message to a channel. Anything other than your own chat or a conversation you approved waits for your OK.
  • Approvals on plugin channels. A channel with its own approval buttons shows them, and Kepler accepts the answer only for a request it asked on that channel.
  • One account, one channel. Connecting a second channel with a bot token already in use is refused.
  • A message that cannot be delivered lands in the Feed instead of being lost. A channel that keeps failing stops and shows why.
Automations
  • Chat trigger. Give an automation a /command and anyone you allowed can run it from a connected chat, or you from the desktop / menu. See Run one from a chat.
  • Post step. Answer in the chat that started the run, send the result to you on a channel, or post to a named conversation, which asks your OK at Go live. See Post in a chat.
  • Decide item by item. A decision step can judge each item in a list, up to 200, in one step.
Also
  • Background lanes over the limit wait in a queue and start in order, instead of being refused. See Specialist agents.
  • The hook log is on the Hooks tab of Activity, under Settings > Plan & Usage.
  • Kepler’s own tools can no longer read or use the key Kepler signs in with.
​
The decision model
  • Decision model. An optional second model, picked under Settings > Models, that answers typed questions in about a second. See Decision model.
    • It double-checks shell commands and MCP calls. It can only make a decision stricter: a confident “write” on a command the rules would have let through asks you instead. Its verdict shows as a pill on the tool row.
    • On the unknown-command card it suggests whether the command reads or writes. You still decide.
    • In chat and war rooms Kepler can use it to classify, score and pick across many items in one call.
    • Automations get a Decide with a decision model step.
    • A watcher that checks for a condition in plain words uses it when one is picked.
  • Run once everywhere. The unknown-command card offers Run once in every posture, Observe included. Watchers and automations still never run a command Kepler cannot vouch for. See The unknown-command card.
  • Reasoning effort is a slider in the model picker. Auto lets the model decide on OpenAI and OpenRouter too.
  • An MCP tool call shows its server’s icon, and its label names the action.
  • A System One endpoint lists the models it serves.
  • A model that can draw images does so without an image model set up.
  • A topology source that fails to scan offers Try again, Ignore and Ask Kepler.
  • A run you cancel still counts the tokens it spent in Plan & Usage.
  • Settings > General > About shows the product version.
​
Remote access and a map that knows more
Work on other machines
  • Remote access. Turn on Work on other machines in Settings > Remote access, then pick a host, a cloud VM, a container or a sandbox from the control beside Send, or type /connect <host>. Kepler connects with your own logins and never asks for or stores a password. See Remote access.
  • Machines come from your SSH config, Google Cloud and AWS, Teleport, Docker, Kubernetes pods, Ansible inventories, and groups made from SSH config patterns. Windows hosts work over SSH.
  • Connected is not the same as allowed. A thread uses a machine only after you pick it or approve it there, and each machine can carry its own limits.
  • Save a machine appends a Host block to your own ~/.ssh/config, after showing it to you.
  • Commands that keep running. Kepler can keep a port-forward or a log tail running for up to twelve hours, with a Stop button. End a shell-mode line with & to do the same yourself. See Commands running in the background.
Topology
  • How things got there. CI workflows from GitHub Actions, GitLab CI, CircleCI and Jenkins, Helm releases, and Argo CD and Flux applications are on the map. See Topology.
  • Where the data lives. Databases, caches, queues and buckets on AWS and Google Cloud, with their addresses. Azure joins the scan, one source per subscription.
  • Where to look. Prometheus, Grafana, Alertmanager, Loki, Jaeger and similar tools found in your clusters, with their addresses.
  • Things no scan can see. Add them to Kepler’s own list, or accept what Kepler proposes from your conversations under Activity > Needs you. See Add what no scan can see.
Also
  • The composer fits a narrow panel, and the conversation’s card stays beside it in narrower windows.
  • Choosing a command that takes an argument puts it in the composer and waits. Tab completes any command and Enter runs it.
  • Rows from a connected Grafana, Datadog, Prometheus or OpenTelemetry integration reach the map.
  • One integration tool with an unusual schema can no longer make every turn fail.
​
Tabs and split panes
  • Tabs. Cmd+T opens a tab, Cmd+1 to Cmd+9 pick one, and Cmd+W closes one. Cmd-click a session in the sidebar, or choose Open in new tab, to open it in a tab. See Tabs and split panes.
  • Split panes. Cmd+D splits right and Cmd+Shift+D splits below. Drag a session onto a pane to place it. Panes you are not typing in keep streaming.
  • Tabs and splits come back after a relaunch.
  • Every shortcut in one place. Press Cmd+/ or type /shortcuts. On Windows and Linux, keys are written as Ctrl, Shift and Alt everywhere in the app. See Commands and shortcuts.
  • Your own messages sit on the right, and code blocks use the editor’s colours.
  • What Kepler can do (/help) has walkthroughs for Automations and for tabs and panes.
  • Ctrl+D splits on Windows and Linux. Cmd+W closes the pane or tab, never the window.
​
Automations
  • Automations have their own place. Describe one, or start from one of 15 templates. The list puts the ones waiting on you first. See Automations.
  • The automation builder. The chat on an automation’s page tries each step before adding it, and cannot run, delete or take an automation live. See The automation builder.
  • Automations can change things, with your permission. Go live lists every step that changes something and asks how each may run: Every run or Ask me each time. See What an automation may change.
  • Event triggers. An automation can start when a check’s result changes, or when a webhook arrives. Webhooks have Send a test and New token. See Webhooks.
  • Approvals and reports in the Feed. Waiting is the one place to answer an approval. Reports can land quietly, and an alert clears itself when a later run finds the problem gone. See The Feed.
  • Integrations stay signed in. Sign-ins renew in the background. See Signing in.
  • Spike and Oodle join the integrations catalog.
  • Sources in web answers show as chips with each site’s icon and name.