curl -fsSL https://raw.githubusercontent.com/bitan-del/gods-eye/main/scripts/install.sh | bash
Gods Eye is a personal AI gateway you run on your own hardware. One command installs it. It connects to 25+ messaging channels (WhatsApp, Telegram, Discord, Slack, Signal, Teams, Matrix, and more), speaks and listens across macOS/iOS/Android, and gives you a single AI brain that remembers everything across every channel.
We studied every pain point users face with other AI tools, researched the best solutions from LiteLLM, RouteLLM, Mem0, LlamaFirewall, Flutter Doctor, Elm, and Factory.ai — then built 24 production modules with 635+ tests to solve all 8 major problems.
Website · Docs · Getting Started · Discord
curl -fsSL https://raw.githubusercontent.com/bitan-del/gods-eye/main/scripts/install.sh | bashThe installer handles everything: Node.js, dependencies, system health checks, and walks you through setup with the quickstart wizard.
Or install manually:
npm install -g godseye@latest
godseye quickstartRuntime: Node 22+ (Node 24 recommended).
We studied every GitHub issue, Reddit complaint, and Discord rant about OpenClaw. Then we researched the best solutions from LiteLLM, RouteLLM, Mem0, LlamaFirewall, Flutter Doctor, Elm, and Factory.ai — and built 24 production modules to fix all 8 pain points.
| # | Problem | OpenClaw (Broken) | Gods Eye (Fixed) | Module |
|---|---|---|---|---|
| 1 | Setup Hell | 47-step manual config, cryptic YAML errors, no validation | godseye quickstart — one-command wizard + Flutter-style godseye doctor health checks |
Doctor Framework, API Key Validation |
| 2 | Cost Explosions | No budget controls, users report $200+ surprise bills, no model routing | LiteLLM-style token budgets (per-session/hour/day) + RouteLLM-style smart routing to cheap models for simple queries | Token Budget Engine, Smart Model Router, Semantic Cache |
| 3 | Context Amnesia | Compaction destroys critical info, AI forgets mid-conversation | Mem0-style pinned memory (persists through compaction) + Factory.ai-style structured JSON compaction with typed sections | Pinned Memory, Structured Compaction, Context Health Monitor |
| 4 | Runaway Agents | Infinite tool loops, no circuit breaker, agents burn tokens doing nothing | 3-state circuit breaker (closed/degraded/open) + progressive autonomy: supervised -> semi-auto -> autonomous | Loop Detection v2, Autonomy Governor, Task Verification |
| 5 | Security Gaps | No input scanning, no prompt injection detection, no output leak checks | Meta LlamaFirewall-inspired 3-layer runtime firewall: input scanning (10 injection patterns + base64), execution auditing, output leak detection | Agent Firewall, Supply Chain Integrity, Network Exposure Audit |
| 6 | Over-Permissioning | Root-level access by default, no permission model, no audit trail | OWASP-aligned permission profiles (minimal/standard/power/unrestricted) + 5-layer audit trail with automatic secret redaction | Permission Profiles, Permission Audit Trail |
| 7 | Not Beginner-Friendly | "Read the docs" is the entire onboarding, error messages are stack traces | Interactive 7-step tutorial, Elm-style friendly errors (E001-E010 with fix hints), 6 ready-to-use agent templates | Interactive Tutorial, Friendly Errors, Agent Templates |
| 8 | Channel Failures | Silent disconnects, no diagnostics, users don't know what's broken | Per-channel health diagnostics with auto-fix hints for Telegram/Discord/Slack/WhatsApp/Signal + unified auth error handler | Channel Diagnostics, Auth Error Handler |
Total: 24 modules. 635+ tests. Every problem solved.
WhatsApp / Telegram / Slack / Discord / Signal / iMessage / Teams / Matrix / 25+ channels
|
v
+-------------------------------+ +---------------------------+
| Gods Eye Gateway | | Gods Eye Online |
| (local control plane) |<--->| (AI Creative Suite) |
| ws://127.0.0.1:18789 | | MCP Bridge :18790 |
+-------------------------------+ +---------------------------+
|
+----------+----------+
| | |
v v v
Pi Agent CLI Tools Apps
(RPC) (godseye) (macOS/iOS/Android)
"If setup is hard, users never get to the good stuff."
- Doctor Framework — Flutter-style system health checks with auto-fix (Node, ports, Docker, disk, config)
- API Key Validation — 3-step validation: format regex, API probe, capability check
- Auto Token Generation — Cryptographically secure gateway tokens (no more "change-me" defaults)
- Network Exposure Audit — Detect dangerous bind configurations and Docker exposure
- Supply Chain Integrity — SHA256 verification + prompt injection detection in MCP tool descriptions
- Agent Firewall — Meta LlamaFirewall-inspired 3-layer runtime protection (input scanning, execution auditing, output leak detection)
"Your AI should work for you, not burn your wallet."
- Token Budget Engine — LiteLLM-style per-session, per-hour, per-day spending caps with configurable warning thresholds
- Smart Model Router — RouteLLM-style complexity classification that routes simple queries to cheap models and complex ones to powerful models
- Semantic Cache — GPTCache-style two-tier caching (exact hash + trigram Jaccard similarity) with LRU eviction
- Loop Detection v2 — 3-state circuit breaker (closed/degraded/open), objective drift detection, and tool repetition analysis
- Autonomy Governor — Progressive trust system: supervised -> semi-autonomous -> autonomous, with automatic escalation at trust score thresholds
- Task Verification — Structured completion checking with accept/retry/escalate recommendations
"An AI that forgets what you told it is an AI that wastes your time."
- Context Health Monitor — Real-time token utilization tracking with warning/critical thresholds and estimated turns remaining
- Pinned Memory — Mem0-style persistent facts that survive compaction, with priority scoring, search, and /remember /forget commands
- Structured Compaction — Factory.ai-style anchored iterative merging using typed JSON sections (not free-text summaries)
- Session Checkpointing — LangGraph-style save/resume/time-travel with full diff support
- Permission Profiles — OWASP-aligned presets (minimal/standard/power/unrestricted) with glob-pattern tool policies and path restrictions
- Permission Audit Trail — 5-layer structured logging (Identity -> Input -> Reasoning -> Execution -> Outcome) with automatic secret redaction
"If your grandma can't set it up, your onboarding is broken."
- Quickstart Command — One-command setup wizard: detects environment, validates API keys in real-time, generates config, starts gateway
- Elm-Style Friendly Errors — Color-coded error catalog (E001-E010) with plain-English problems and actionable fix hints
- Agent Templates — 6 ready-to-use configs: code-reviewer, research-assistant, devops-automator, content-writer, data-analyst, security-auditor
- Interactive Tutorial — 7-step guided walkthrough with progress tracking and completion percentage
- Channel Diagnostics — Per-channel health checks for Telegram, Discord, Slack, WhatsApp, Signal — with auto-fix suggestions
- Unified Auth Error Handler — Provider-specific error messages with remediation links and exponential backoff retry logic
# Install
curl -fsSL https://raw.githubusercontent.com/bitan-del/gods-eye/main/scripts/install.sh | bash
# Or if already installed:
godseye quickstart # guided setup wizard
godseye doctor # system health check
godseye tutorial # interactive walkthrough
# Start the gateway
godseye gateway --port 18789 --verbose
# Talk to your assistant
godseye agent --message "Hello from Gods Eye" --thinking high
# Send a message via any channel
godseye message send --to +1234567890 --message "Hello from Gods Eye"The unified creative brain for your assistant — image generation, video generation, brand intelligence, and content calendar management.
godseye agent --message "generate a hero image for my landing page"
godseye agent --message "scan my brand from https://example.com"
godseye agent --message "show my content calendar for this week"BYOK (bring your own keys): Studio uses your API keys for fal.ai, Google Gemini, and OpenAI. Set them during godseye quickstart or via godseye config set.
Docs: Getting started · Brain architecture · Creative providers
The cloud-hosted creative suite that connects to your local Gods Eye gateway via MCP bridge:
- Image Generation — Gemini Flash, Flux Ultra, SeedDream v4.5
- Video Generation — Kling 3 Pro, Veo 3.1 Fast
- Chat Intelligence — Gemini 3.x, Kimi K2.5
- Brand DNA Scanner — Extract colors, fonts, tone from any website
- Content Calendar — Plan and schedule across 10+ social platforms
Connect your local gateway to Gods Eye Online:
godseye quickstart # sets up the MCP bridge automaticallyWhatsApp, Telegram, Slack, Discord, Google Chat, Signal, BlueBubbles (iMessage), iMessage (legacy), IRC, Microsoft Teams, Matrix, Feishu, LINE, Mattermost, Nextcloud Talk, Nostr, Synology Chat, Tlon, Twitch, Zalo, Zalo Personal, WeChat, WebChat, macOS, iOS, Android.
Every channel gets per-channel diagnostics:
godseye doctor # checks all connected channelsgit clone https://github.com/bitan-del/gods-eye.git
cd gods-eye
pnpm install
pnpm build
pnpm godseye quickstart
# Dev loop (auto-reload)
pnpm gateway:watch
# Run tests (635+ tests)
pnpm test- macOS App — Menu bar control plane, Voice Wake/PTT, Talk Mode overlay, WebChat, debug tools
- iOS Node — Canvas, Voice Wake, Talk Mode, camera, screen recording, Bonjour pairing
- Android Node — Chat, voice, Canvas, camera/screen recording, device commands
Gods Eye treats security as a first-class citizen, not an afterthought:
- 3-layer Agent Firewall — Input scanning (10 injection patterns + base64 detection), execution auditing (privilege escalation, exfiltration, dangerous sequences), output scanning (API key leaks, sensitive paths, XSS)
- Supply Chain Verification — SHA256 integrity checks + prompt injection detection in MCP tool descriptions
- Permission Profiles — OWASP-aligned presets so you never over-permission your agent
- 5-layer Audit Trail — Every action logged with automatic secret redaction
- Network Exposure Audit — Detects risky bind configurations before they become breaches
- DM Pairing — Unknown senders get a pairing code; no message processing until approved
Full security guide: Security
"The best AI infrastructure doesn't have to come from Silicon Valley."
Gods Eye is proudly engineered in India. 24 modules, 635+ tests, 12,000+ lines of production code — every line written with the conviction that world-class AI tooling can come from anywhere.
We didn't fork. We didn't patch. We studied the problems, researched the best solutions from across the industry, and built something better from scratch.
Built with intensity. Built to be unstoppable.
See CONTRIBUTING.md for guidelines.
GODS EYE
Built with intensity in India by @bitan-del
24 modules · 635+ tests · 25+ channels · zero compromises