A desktop dashboard for managing Claude agent sessions.
Less noise. More visibility. Full control.
Otto is a desktop dashboard for professional developers who run multiple Claude agents simultaneously. It replaces noisy terminal logs with a clean, real-time overview — session status, file diffs, costs, and tasks all visible at a glance. Less scrolling, more doing.
It's built around GitHub. Launch agents from issues, PRs, and discussions, manage worktrees, and automate sessions with webhooks and cron triggers — so your agents can respond to events without you being there.
Otto works with the Anthropic API, Amazon Bedrock, Google Vertex AI, and Azure AI Foundry. If you're on Bedrock, Vertex, or Azure and don't have access to Claude's native apps, Otto gives you a first-class experience regardless of your provider.
- Launch and monitor multiple agent sessions simultaneously from a single dashboard
- Stage overview with tiered session grouping: Needs Attention, Working, Idle, and Inactive
- Stage cards show a thumbnail of the agent's most recent browser capture, so you can spot who's driving a page without opening each session
- Sessions requiring human input are surfaced at the top automatically
- Session pinning, forking, renaming, and history replay
- Filter and search across all sessions
- Agent orchestration — agents can spawn child sessions to delegate work and send messages between sessions to coordinate
- Live streaming transcripts with rich content types — code, files, tool calls, thinking blocks, memory recalls
- "Now Playing" header showing the agent's current activity at a glance
- File checkpoint and rewind system with visual diff preview before confirming
- Transcript annotations — highlight text and attach comments as context for the next prompt
- Default view (the Handover) — the session view built for the moment you return to a session. Top to bottom it answers why you're here, what happened, and what's being asked, and lets you act without leaving the page. A state banner tells you the session's phase at a glance (needs your decision, working, turn finished, or failed); permission, question, and plan prompts render inline as the primary element instead of popping up as separate dialogs; the agent's latest message sits front and centre with controls to cycle back through earlier updates — highlight any passage in it to pin a note beside that line, which carries into your next reply; and a receipts row shows outcome-shaped evidence — files changed with diffstats you can click straight through to, command pass/fail counts, and task progress; when the agent delegates work to sub-agents, they show up as live activity strips right above where you reply, collapsing to a compact list once finished
- Four more session views to switch to whenever you want a different lens: the classic Dashboard, plus Neural Pulse, Mission Timeline, and Cockpit HUD
- Neural Pulse — a living constellation of the session: each prompt anchors a hub, with its thinking and tool calls orbiting it, and orb size scales with token usage (sub-agent orbs grow live as they work). Interactive, zoomable and pannable, with sub-agents as integrated branches
- Mission Timeline — mission control layout with parallel activity tracks, draggable minimap, and click-to-inspect detail panel
- Cockpit HUD — flight instrument aesthetic with context gauge, tool radar, live cost and token gauges, and transcript message cycling
- View picker toggle with persistent selection across sessions
- Granular session configuration: permission modes, thinking budgets, cost limits, turn limits, sandbox mode, tool restrictions, structured output, and more
- Interactive permission and plan approval with visual previews
- Auto-approval mode — safe tool requests are approved automatically; risky or uncertain requests still surface for manual review, and any request the automatic check blocks is recorded in the transcript so you can see what was denied and why
- Follow-up message queueing while agents are working
- Prompt suggestions, slash command autocomplete, and drag-and-drop file attachments — any file type, not just images; Otto reads the file where it lives, no copying required
- Provider profiles — configure and switch between API providers, with environment isolation and auto-detection of existing credentials
- Change model, permission mode, reasoning effort, and thinking on/off mid-session
- Review a plan right where you read it — highlight any passage to leave a note pinned beside it, instead of a separate review dialog. When approving, pick the implementation permission mode, reasoning effort, and (optionally) a different model — so you can plan with Opus and build with Sonnet on a fresh context window. Clear context is also available mid-session from the sidebar.
- When Claude presents options or asks questions, Otto renders fully interactive HTML previews directly in the dashboard
- See UI mockups, architecture diagrams, and design options as actual rendered visuals — not markdown descriptions
- Options and their preview sit side by side, so you can compare choices and the mockup at once without scrolling
- Quickly verify your agent understands the task before it starts building
- Agents can open web pages directly inside the session panel for you to review — no separate window needed
- Resizable browser panel with device viewport presets (mobile, tablet, desktop), fullscreen expand, and toolbar controls
- Annotation overlay lets you highlight elements on the page, add comments, and send structured feedback back to the agent
- Element screenshots and selectors are included in the feedback so the agent knows exactly what you're referring to
- Agents can browse autonomously — when an agent drives the browser itself, the panel minimises to an "Agent browsing" indicator you can expand to watch the page live, then collapse again when you're done
- The Default view's receipts row shows a browser card whenever an agent has a page open — page title, status, and a live preview of its last capture, with arrows to look back through recent screenshots
- Worktree overview showing linked agent sessions per worktree
- GitHub panel with Issues, Pull Requests, Discussions, CI Runs, and Tags
- One-click "Launch Agent" from any GitHub item — automatically creates a worktree
- Full worktree lifecycle management with batch operations and safety checks
- Self-assignment on issues, label filtering, fuzzy search
- Conflicted PRs are flagged in the list, with a one-click Resolve conflicts action that spins up a worktree agent to merge the base branch in safely (no force-pushes)
- Branch changes panel — a live diffstat of your branch against main, with a slide-over showing branch and push state, your linked pull request (review and CI status), the branch's own commits, and every changed file grouped as Staged / Unstaged / Untracked — click any file for its diff. Commits expand in place to show the files they touched, each opening its diff at that commit. Available from every session view: as a tile in Default and Dashboard, or via the sidebar's file-count stat elsewhere
- Dedicated Automations tab with structured editors for cron schedules and GitHub webhook rules — no need to hand-edit files
- Cron editor with schedule presets and a live next-fires preview, so you can see when your automation will actually run
- Webhook match builder with event/action chips, repo/sender/label filters, and payload matchers — plus a Form/Raw toggle for power users
- Live event feed and Insights panel showing top event types, recent volume patterns, and one-click "Create automation" suggestions for frequent events
- Per-automation session configuration (model, reasoning effort, permissions, worktree, sandbox, budget, and more)
- Per-repo on/off toggles in the header for crons and webhooks — automations are off by default for each repo, so nothing fires until you explicitly enable it
- Dedicated Hooks tab with structured editors for command, prompt, agent, and HTTP hooks across user, project, and local settings
- Test panel — run any hook against a synthesised payload, see exit code, stdout, and stderr without leaving the dashboard
- Recent-fires strip per hook showing the last execution outcomes from live sessions, so you can spot misbehaving hooks at a glance
- Enable or disable hooks without deleting them — Otto stashes disabled hooks safely so Claude never sees them
- Inline rename with Otto-managed metadata, plus a Form/Raw toggle for power users
- Dedicated Memory tab unifying both Claude memory systems — Instructions (CLAUDE.md hierarchy) and Learned (auto-memory) — in one place
- View, edit, create, and delete entries at every level you own, across user and project scopes
- Header stat line shows entry counts and approximate token usage at a glance
- Memory behaviour settings tucked into a collapsible section so the main view stays focused on content
- Dream now button opens a pre-filled session that reviews and consolidates your learned memories on demand
- Auto-dream toggle schedules nightly consolidation itself, so it works reliably on every supported provider — not just when signed in directly with Claude
- Project picker with automatic git repository scanning
- Quick-switch navigation between repositories
- Multi-repo projects — a single project can contain multiple git repositories
- Drag to reorder project tabs in the header — your preferred ordering is remembered across reloads
- Unified browse view across all configured marketplaces — installed and available plugins side by side, no dropdown prerequisite
- See what each plugin ships before installing — skill, agent, command, hook, MCP, and LSP counts on every card, with a full named breakdown and descriptions on the detail page
- Dedicated plugin detail pages with author, repository, licence, and keywords surfaced from marketplace manifests
- Enable, disable, and configure plugins at user, project, or local level — inline from the detail page
- Add marketplace sources from GitHub, npm, git, URL, file, or directory
- Trust and security controls: strict mode, blocked sources, and custom trust messages
- In-app feedback tool — annotate UI elements and submit bug reports directly from the dashboard
- Automatic updates with in-app download and restart
- Detailed context usage breakdown with token counts by category
- Live cost and token tracking across sessions, with a cost detail view showing total tokens, cache hit rate, and exact counts — spend survives reconnects, reloads, and restarts
- SDK recording and debugging tools
- Mobile-responsive layout
Coming soon. Otto is currently in development. Watch this repo to be notified when downloads are available.
- Download and install Otto from the links above
- Open Otto — the dashboard opens automatically
- Select a project — Otto scans for git repositories and lets you pick one
- Create a session — configure permissions, thinking budget, and tools, then launch your first agent
- Watch it work — follow the real-time transcript, approve permissions when asked, and queue follow-up messages
Otto is currently in closed beta. You can download and install it, but you'll need a licence key to use it.
If you'd like to try Otto, open a discussion or reach out directly to request access.
Note: Otto is the dashboard. You bring your own API access — Anthropic API, Amazon Bedrock, Google Vertex AI, or Azure AI Foundry.
- macOS 12 or later (Apple Silicon or Intel)
- One of the following:
- An Anthropic API key
- Amazon Bedrock access with Claude enabled
- Google Vertex AI access with Claude enabled
- Azure AI Foundry access with Claude enabled
- Bug reports — Open an issue
- Feature requests — Open an issue
- Questions & discussion — Start a discussion
Made by m2de