What is Companion OS?
Companion OS is an embodied AI platform. Your companion lives in three places simultaneously — your Ray-Ban Meta glasses, your phone screen, and a desktop agent that works for you 24/7. It's not a chatbot. It's a presence.
Glasses
Hears and sees everything through your Ray-Ban Meta. Answers hands-free.
Phone
3D avatar on screen. Speaks, listens, and remembers every session.
Desktop
Runs autonomous tasks — email, research, goals — while you sleep.
Setup guide
Download the app
Get the Android APK from the landing page, or install the Windows desktop package. iOS builds ship in v2 after App Store review.
# Android Download companion-os-v1.apk from companion-os.xyz Settings → Install unknown apps → enable for your browser Open the APK → Install # Windows Download CompanionOS-Setup.exe Run installer → Launch Companion OS
Onboarding flow
5-step setup when you first launch:
- Choose archetype — Nova, Aria, Orion, Vex, Kira, or Create your own (custom keywords + tone)
- Name your companion
- Integrations — connect Telegram, email, calendar (optional)
- Age verification
- Character Studio — Face Scan / Face & Body / Hair / Style / Outfit / Import VRM (skippable, uses archetype preset if skipped)
Connect Ray-Ban glasses (optional)
Companion OS connects to Ray-Ban Meta glasses via WebSocket. The companion sees what you see and hears through the glasses mic.
1. Pair glasses in Meta View app 2. In Companion OS onboarding step 5: - Enter glasses IP (found in Meta View → Developer settings) - Enter stream port (default: 9870) 3. Status indicator shows "👓 Glasses" when connected Falls back to "📱 Phone camera" automatically
Import memory (optional)
Paste a conversation export from ChatGPT or Claude. Companion OS extracts key facts about you and builds the initial life graph without starting from zero.
ChatGPT: Settings → Data controls → Export data → conversations.json
Claude: claude.ai → Profile → Export data
The semantic memory layer
The Companion Twin is a persistent world model of the user — a graph of people, projects, habits, goals, and events that grows with every session.
Life Graph
liveEntities (people, projects, places, goals, habits, tokens) connected by typed relationships. Extracted from every session via Gemini. Injected into every system prompt so the companion always knows your world.
// Entity types person | project | place | concept | habit goal | event | tool | token | org // Example triples user → works_on → YourProject user → was_seen_at → coffee shop user → engages_in → coding
Predictive Activation
livePre-emptive intelligence. Generates morning brief cards, goal nudge cards, and pre-meeting context automatically. Cards surface in the chat interface when relevant.
Morning brief
3 priorities, market snapshot, calendar
Goal nudge
Triggered when a goal has no activity for 3+ days
Meeting context
Who you last spoke to, what was discussed
Contextual Vision
liveCamera frame → Gemini Vision describe → extract entities → store → feed to Life Graph. Works via glasses camera or phone camera.
Frame pipeline:
1. Capture frame (glasses or phone)
2. Describe: "Person is working at a desk, whiteboard visible"
3. Extract: { location: "home desk", activity: "coding", significance: 0.7 }
4. Store in vision_memory.json
5. If significance ≥ 0.5 → feed entities to Life Graph
6. Inject recent observations into every sessionGoal Drift Detector
liveAudits GOALS.md against recent session logs every 6 hours. Fires a nudge card when a goal hasn't been mentioned in 3+ days. 2-day cooldown per goal.
Skills from natural language
The companion can create, save, and run reusable skills from plain English. Say what you want once — it becomes a named, callable workflow.
Skill types
prompt-skillSingle-shot: system prompt + tools + output. Best for analysis, briefings, lookups.
workflow-skillMulti-step tool chain. Steps execute sequentially, pause for confirmation where needed.
advisor-skillAnalyze and recommend only. Never acts without explicit approval. Default for money/trades.
Creating skills
// Voice or chat "Make a skill that scans new Solana launches for rug risk" "Whenever I say meeting prep, pull Gmail + calendar + contacts" "Create a skill to summarize my unread emails every morning" "Save this workflow as a reusable skill" // Updates "Update that skill to also check holder concentration" "Rename it to Launch Filter" "Disable that skill" "Make it approval-only" "Add a voice trigger: token check"
Seeded skills (available out of the box)
Analyzes a token launch for rug risk via DexScreener + on-chain signals. Scores 0–10.
Pulls Gmail + calendar + life graph context → pre-meeting brief in seconds.
Email + calendar + market snapshot → prioritized daily brief.
Audits goals vs recent activity. Flags what's stalled, what's on track.
Web search + memory → 300-word research brief on any topic.
Advanced twin modules
Market Brain
The Solana operator console. Live token prices via Jupiter API, DexScreener signals, Gemini-powered memecoin risk scoring, Solana narrative detection via Google Search grounding.
// Example: memecoin risk score
{
address: "ABC...XYZ",
symbol: "LAUNCH",
liquidityUsd: 45000,
holderConcentration: 0.72, // top wallet holds 72%
rugRisk: "HIGH",
flags: ["low-liquidity", "whale-concentration", "new-contract"]
}Outcome Twin
Personal world simulator. Reasons over your actual goals and life graph to model the downstream effects of a decision before you make it.
"If I work on the side project tonight instead of the client deadline, what slips?"
→ immediateEffects: ["Client deadline at risk", ...]
→ goalImpact: [{ goal: "client delivery", impact: "negative", ... }]
→ recommendation: "Side project can wait 48h. Client deadline can't."
"If I enter this trade, what does that do to my risk budget?"
→ tradeoffs: ["$X of liquidity locked", "correlates with SOL position", ...]Counterparty Twin
Trust graph for wallets, founders, protocols, and people. Not just “what is this token” — “should I trust this actor?”
// Trust score: 0 (confirmed bad) → 1 (verified trustworthy)
{
identifier: "7abc...def1",
type: "wallet",
trustScore: 0.23,
tags: ["rug-history", "low-liquidity", "new-wallet"],
signals: [
{ type: "on-chain", sentiment: "negative", weight: 0.8,
description: "Associated with 3 previous rug exits" }
]
}Narrative Twin
Memetic attention radar. Finds what's heating up in Web3 before the price move is obvious. Powered by Gemini Search grounding — real-time signal, not cached data.
// Attention radar output Heating: "AI + DeFi convergence" [heat: 8.1/10 ↑] Emerging: "RWA Season 2" [heat: 5.4/10 ↑↑] Cooling: "L2 fee wars" [heat: 3.2/10 ↓] // Early signals (pre-price-move candidates) → high velocity + emerging status = watch closely
Embodied vision + voice
Ray-Ban Meta glasses are what make Companion OS genuinely embodied. Your companion hears, sees, and speaks through the glasses — hands-free, eyes-free, real-world aware.
Vision pipeline
Camera frames → Gemini Vision → entity extraction → Life Graph. The companion builds visual memory of your world automatically.
Voice pipeline
Glasses mic → transcription → Gemini → response → glasses speaker. Full conversation without touching your phone.
Burst capture
Not every frame — burst-based on configurable interval or significant scene change. Keeps API costs low while building rich visual memory.
Phone fallback
No glasses? Companion OS falls back to phone camera automatically. The same vision pipeline runs through the phone camera.
// GlassesStreamService WebSocket protocol
ws://{glassesIp}:{port} → JPEG frames
CompanionClaw receives frame → VisionMemory.processFrame()
→ Gemini Vision describes: "User at coffee shop, MacBook open, latte"
→ Entities extracted: { location: "coffee shop", activity: "working" }
→ Life Graph updated
→ Next session: companion knows you work from Brew & Co on TuesdaysThe intelligence backend
CompanionClaw is the multi-tenant backend that powers Companion OS. Every user gets a fully isolated workspace with self-heal, self-learn, cron jobs, memory, and skill execution.
Cron schedule (per user)
| Job | Schedule | What it does |
|---|---|---|
| morning-brief | 8am daily | Goals + market snapshot + narrative radar → push notification |
| task-executor | Every 2h | Executes highest-priority task in TASKS.md |
| memory-reconcile | 3am daily | Distills session logs → structured memory + Life Graph extraction |
| task-planner | 7am Monday | Reviews GOALS.md → seeds 3-5 tasks for the week |
| intel-sweep | 11am daily | Web search on active goals → briefing |
| goal-auditor | Every 6h | GoalDriftDetector audit → nudge cards if stalled |
| weekly-review | 6pm Sunday | Wins, gaps, next week focus → push notification |
| nightly-check | 9pm daily | SkillWatcher scan + Narrative Twin refresh |
Self-heal
HeartbeatMonitor runs every 5 minutes per active session. Checks gateway health, Gemini session, disk. Restarts and notifies on repeated failures.
Memory reconciliation
// 3am daily per user 1. Read all session logs from past 24h 2. Gemini extracts: new facts, updated preferences, resolved tasks 3. Merge into memory.json with confidence scores 4. Update USER.md and GOALS.md if new goals detected 5. Extract Life Graph entities from session log 6. Regenerate memory_for_model.json (flat projection for session injection)
Utility and tiers
$COMPANION is the Solana token that gates access tiers, powers the VRM marketplace, and aligns builders with the platform's growth.
Free
0 $COMPANION
- ✓ 3 preset avatars
- ✓ 7-day memory
- ✓ 1 companion
- ✓ Managed API key (30 min/day)
Holder
≥100 $COMPANION
- ✓ Full VRM library
- ✓ 90-day memory
- ✓ VRoid Hub import
- ✓ Managed key (3h/day)
Staker
≥1,000 $COMPANION
- ✓ Unlimited memory
- ✓ 3 companions
- ✓ Early access
- ✓ BYOK support
Builder
≥10,000 $COMPANION
- ✓ API access
- ✓ Host for others
- ✓ Referral revenue
- ✓ Multi-agent swarms
NFT Companion Archetypes
10 premium companions with unique personalities and rare VRMs minted as Solana NFTs. Holders earn $COMPANION when others run their archetype.
VRM Marketplace
Creators list custom 3D avatars priced in $COMPANION. 5% platform fee to treasury. VRoid Hub import free for first 90 days.
What's next
Live now
- Web demo (companion-web.vercel.app)
- VRM avatar system — FBX idle animation, lip sync
- Face scan → Gemini anime portrait + color match
- Character Studio: Face Scan / Face & Body / Hair / Style / Outfit / Import VRM
- VRoid Hub OAuth import
- Custom companion persona creator
- Privacy Center (Wave 1–3)
- Image generation (Imagen 4)
- Ray-Ban glasses streaming (VisionClaw)
- Gemini web search + vision in chat
Pre-launch
- CompanionClaw Railway deploy (backend ready, deploy pending)
- Companion Twin V1–V3 + V5
- Skill Forge
- Morning brief + push delivery
- Stable APK build
- Windows desktop package
- Video generation via Veo 3
- Email + calendar UI
- $COMPANION token launch (Pump.fun)
v2
- iOS App Store submission
- Multi-agent swarms (Staker+ tier)
- Computer use agent (Builder tier)
- Voice clone
- Twitter @CompanionOS archetypes
- Play Store submission
- Skill marketplace
v3
- Unity bridge (console-quality avatar rendering)
- On-device memory (no cloud for privacy tier)
- Companion-to-companion social graph
- Group companion sessions
- Developer SDK
CompanionClaw REST API
CompanionClaw endpoints are hosted at https://api.companion-os.xyz (or your Railway URL). Pass x-gemini-key header for BYOK.
Companion-web also exposes API routes at /api/* for the demo layer: /api/demo-chat, /api/avatar/stylize, /api/generate-persona, /api/vroid/* (OAuth + model browse + download).
/healthGateway health check
/onboardProvision a new user workspace from onboarding data
/chatSend a chat message. Detects forge intent and skill triggers automatically.
/users/:id/twinGet life graph context block + goal drift summary
/users/:id/twin/extractTrigger manual life graph extraction from text
/users/:id/activationGet pending proactive activation cards
/users/:id/skills/forgeCreate a skill from a natural language prompt
/users/:id/skillsList all enabled skills
/users/:id/skills/:skillId/runExecute a skill by ID
/users/:id/skills/:skillIdUpdate a skill via NL or direct patch
/users/:id/skills/:skillIdDisable or archive a skill
/users/:id/outcome/simulateSimulate the downstream effects of a decision
/users/:id/counterparty/analyzeAnalyze a wallet, founder, or protocol for trust signals
/users/:id/narrative/radarGet current narrative attention signals
/users/:id/captureStore a vision frame (glasses/phone/webcam) into VisionMemory + Life Graph
/users/:id/avatar/configGet saved avatar config (track, VRM URL, colors)
/users/:id/avatar/configSave avatar config after creator flow
/users/:id/privacy/settingsGet privacy settings (retention, provider-lane, redaction flags)
/users/:id/privacy/settingsUpdate privacy settings
/users/:id/privacy/exportRequest full data export
/users/:id/privacy/deleteDelete account and all associated data
/users/:id/privacy/auditGet audit trail of AI calls and data access events
/market/briefSolana market brief (no userId needed)
/market/scoreMemecoin risk score for a contract address
/swarmMulti-agent swarm for complex tasks (Staker+)