Apollo

Apollo window & context injection

A floating, always-on-top window opened from the titlebar Apollo pill. A lightweight, always-available assistant that knows your projects live via Orbit’s MCP server — on open it loads the AI reference, recent completions, and the active queue. The Observatory visual system shows live status pills. Covered by the Sivants subscription.

MCP tool: orbit_chat_send (Apollo is also an MCP client).

Apollo Action system

Inline in the Apollo transcript as status, result, and confirmation cards. Apollo executes real Orbit operations — generate image, read file, queue prompt, open panel, embed icons, run git via orbit_exec_shell — through intercepted [ACTION:...] markers, with confirmation cards for high-complexity or destructive actions. Action cards that complete successfully or expire (because the triggering condition is no longer relevant) auto-resolve to a compact done state; you can also dismiss any action card manually.

Apollo Preferences & Memory panels

Apollo header → Preferences and Memory buttons. Inline panels that open inside the Apollo window without navigating away.

Preferences covers:

• Model selection — concrete model variants, grouped by subscription-included and BYOK (brings-your-own-key) options; adding a missing BYOK key links to the Vault
• Personality — choose from several personality profiles that shape Apollo’s tone and approach
• Thinking level — controls how much chain-of-thought reasoning Apollo applies (None / Minimal / Low / Medium / High)
• Event card visibility — toggle which lifecycle events appear inline in the conversation
• Queue-list collapse preference — whether the active queue collapses in the Apollo context rail by default

Memory shows Apollo’s saved memory as an editable list. Add or delete memory entries directly from the panel; the active set is injected into every Apollo context window.

Apollo Voice

Apollo footer mic button; also configurable from Preferences → Voice.

Two modes: Push-to-talk (hold the mic button or a configured keyboard key to record; release to transcribe and send) and Continuous (hands-free mode that listens until you barge in or stop it). Transcription uses the Web Speech API (“Standard”) or Whisper (“High quality”), selectable in Preferences.

Spoken replies are optional. When enabled, Apollo reads its response aloud using ElevenLabs (James voice by default) or the system’s built-in voice if ElevenLabs is unavailable. Verbosity and the PTT keyboard binding are configurable. Voice input and spoken replies work on mobile as well.

Apollo Chat Attachments

Apollo chat input. Drop or paste up to four image or PDF files into the chat input before sending. Each attachment appears as a removable preview card above the input — images show a thumbnail with a lightbox click, PDFs show a file icon. The input blocks submission while files are still encoding. A warning appears if the currently selected model does not support attachments; Gemini models are required for attachment-capable turns. Files up to 20 MB each are accepted.

Apollo conversation sidebar, persistence & handoff

The Apollo left sidebar, a “Hand off” button, and a context-fill bar. Server-side SQLite conversation history that survives restart, with search, star, rename, and folder organization; persisted collapse state, a provider/model footer, a context-fill meter, and a handoff flow that generates a Flash recap for a fresh conversation.

Apollo Artifacts panel

The Apollo right rail (Ctrl/Cmd+Shift+A). A collapsible panel holding Apollo’s generated artifacts — a live setup checklist, auto-extracted long-reply documents (drag out as .md), and generated images (preview, save to project).

Apollo event notification cards

Inline in the Apollo conversation. Display-only lifecycle cards (build, server, sandbox, prompt events) that surface live or batched per project, never trigger an AI turn, and persist with the conversation.