Skip to content

OPENCLAW_FEATURE_AUDIT.md

Latest commit

 

History

History
864 lines (690 loc) · 37.5 KB

OPENCLAW_FEATURE_AUDIT.md

File metadata and controls

864 lines (690 loc) · 37.5 KB

OpenClaw Feature Audit — Exhaustive Inventory

Date: 2026-03-07 Purpose: Competitive intelligence for Cato development Confidence: 90% — based on official docs, GitHub README, community sources

Table of Contents

  1. Background & Naming History
  2. Architecture Overview
  3. Web Control UI (Browser Dashboard)
  4. WebChat (Native Chat Interface)
  5. Canvas / A2UI (Agent-Driven Visual Workspace)
  6. Messaging Channels (24 Total)
  7. Model Providers & Routing
  8. Memory System
  9. Skills / Plugin System (ClawHub)
  10. Session Management
  11. Cron Jobs & Scheduling
  12. Browser Automation
  13. Exec Approvals & Permissions
  14. Nodes (Companion Devices)
  15. Voice & Audio
  16. Cost / Budget Tracking
  17. Security & Secrets Management
  18. MCP (Model Context Protocol) Integration
  19. Desktop Apps
  20. Agent-to-Agent Communication
  21. Chat Commands
  22. Deployment & Installation
  23. Localization
  24. Known Weaknesses & Community Complaints
  25. Cato vs OpenClaw Gap Analysis

1. Background & Naming History

Name Period Notes
ClawdBot Original Anthropic raised trademark concerns
MoltBot Interim Temporary rebrand
OpenClaw Current (2025+) Final name, open-source
  • Tagline: "Your own personal AI assistant. Any OS. Any Platform."
  • License: Open-source (GitHub: openclaw/openclaw)
  • Written in: TypeScript/Node.js (gateway), Swift (macOS/iOS), Kotlin (Android)
  • Package manager: npm (openclaw package)
  • Stars/Community: 13,729+ skills on ClawHub, very active GitHub Discussions

Status: CONFIRMED

2. Architecture Overview

Core Components

Component Description
Gateway Always-on WebSocket control plane at ws://127.0.0.1:18789
Agent Runtime Assembles context, invokes LLM, executes tools, persists state
Nodes Companion devices (macOS/iOS/Android/headless) connecting via WebSocket
Skills Modular tool bundles (SKILL.md + supporting files)
Canvas Agent-driven visual workspace (separate process, port 18793)
Control UI Browser dashboard served by Gateway at port 18789
WebChat Native chat interface via Gateway WebSocket

Data Flow

User Message (any channel)
  --> Gateway (routing, auth, session lookup)
    --> Agent Runtime (context assembly, LLM call, tool dispatch)
      --> Response back through originating channel

Storage Model

  • Config: ~/.openclaw/openclaw.json (YAML/JSON)
  • Memory: Plain Markdown files on local filesystem
  • Sessions: JSONL transcripts at ~/.openclaw/agents/<agentId>/sessions/<SessionId>.jsonl
  • Search Index: Per-agent SQLite at ~/.openclaw/memory/<agentId>.sqlite
  • Exec Approvals: ~/.openclaw/exec-approvals.json

Status: CONFIRMED

3. Web Control UI (Browser Dashboard)

The Control UI is a browser-based admin surface served at http://127.0.0.1:18789/ by the Gateway.

3.1 Chat Panel

Feature Description Status
Chat with agent Full chat via Gateway WebSocket CONFIRMED
Streaming responses Real-time token streaming CONFIRMED
Tool call cards Live tool output displayed as cards during agent events CONFIRMED
Chat history chat.history retrieval (truncates heavy metadata) CONFIRMED
Chat send chat.send with idempotent messaging CONFIRMED
Chat inject Append assistant notes to transcript without agent run CONFIRMED
Chat abort Stop active runs (/stop, abort phrases) with partial text retention CONFIRMED
Abort metadata Transcript marks partial completions CONFIRMED

3.2 Sessions Panel

Feature Description Status
Sessions list View all active sessions CONFIRMED
Per-session overrides Thinking/verbose toggles per session CONFIRMED
Presence list Show active connections (instances) CONFIRMED
Session refresh Refresh session state CONFIRMED

3.3 Channel Management

Feature Description Status
Channel status Monitor WhatsApp, Telegram, Discord, Slack, etc. CONFIRMED
Plugin channels Mattermost and custom integrations CONFIRMED
QR login QR code capability for channel connections CONFIRMED
Per-channel config Individual channel configuration CONFIRMED
Delivery modes Announce summary, none, or webhook CONFIRMED

3.4 Cron Jobs Panel

Feature Description Status
Job list View all scheduled jobs CONFIRMED
Add/edit/delete jobs Full CRUD for cron jobs CONFIRMED
Run manually Trigger job on-demand CONFIRMED
Enable/disable toggles Per-job activation CONFIRMED
Run history View past execution results CONFIRMED
Delivery modes Announce (default), none, webhook CONFIRMED
Webhook config Bearer token support (cron.webhookToken) CONFIRMED
Delete-after-run One-shot job support CONFIRMED
Agent model overrides Per-job model selection CONFIRMED
Cron stagger Prevent simultaneous execution CONFIRMED
Form validation Field-level error messaging CONFIRMED

3.5 Skills Panel

Feature Description Status
Skills list View all installed skills CONFIRMED
Status display Show enabled/disabled state CONFIRMED
Enable/disable toggle Per-skill activation CONFIRMED
Install new skills Add skills from ClawHub or workspace CONFIRMED
API key management Per-skill API key updates CONFIRMED

3.6 Nodes Panel

Feature Description Status
Node listing View connected companion devices CONFIRMED
Capabilities display Show what each node can do CONFIRMED
Exec approvals card Edit allowlists per node CONFIRMED

3.7 Configuration Panel

Feature Description Status
View/edit config ~/.openclaw/openclaw.json via config.get/config.set CONFIRMED
Apply and restart Config validation + gateway restart CONFIRMED
Base-hash guard Prevents concurrent edit conflicts CONFIRMED
Schema-driven forms Dynamic form rendering for plugins/channels CONFIRMED
Raw JSON editor Fallback for advanced editing CONFIRMED

3.8 Diagnostics & Monitoring

Feature Description Status
Status/health snapshots System health overview CONFIRMED
Models listing Current model state CONFIRMED
Event log Gateway event display CONFIRMED
Live log tail Gateway log with filtering and export CONFIRMED
Debug panel Development diagnostics CONFIRMED
Manual RPC calls Test individual RPC endpoints CONFIRMED
Update management Package/git updates with restart reporting CONFIRMED

3.9 Dashboard Metrics

Metric Description Status
Gateway status Online/offline indicator CONFIRMED
Uptime Time since last restart CONFIRMED
Memory usage RAM consumption CONFIRMED
CPU usage Processor load CONFIRMED
Connected channels Count of active channels CONFIRMED
Message counts Total messages processed CONFIRMED
Error rates Failure frequency CONFIRMED
Last activity Timestamp of most recent event CONFIRMED
Messages per day Daily throughput chart CONFIRMED
Response times Latency metrics CONFIRMED
Token usage Input/output token counts CONFIRMED
Model distribution Which models are used and how often CONFIRMED
Active sessions Current session count CONFIRMED
Session duration Average/per-session timing CONFIRMED
User activity Per-user engagement metrics CONFIRMED

3.10 Authentication & Access

Feature Description Status
Token auth connect.params.auth.token at WebSocket handshake CONFIRMED
Password auth In-memory only, not persisted CONFIRMED
Device pairing First-time browser connection workflow CONFIRMED
Device approval/revocation Via CLI CONFIRMED
Tailscale integration Auto-approval for local, identity header verification CONFIRMED
Loopback auto-approval 127.0.0.1 connections auto-approved CONFIRMED
Device ID generation Per-browser profile CONFIRMED
URL token stripping Tokens removed from URL after load CONFIRMED
Emergency bypass dangerouslyDisableDeviceAuth flag CONFIRMED
Allowed origins CORS configuration for remote deployments CONFIRMED

Status: ALL CONFIRMED from official documentation

4. WebChat (Native Chat Interface)

Feature Description Status
Native WS connection Direct Gateway WebSocket (no embedded browser) CONFIRMED
Deterministic routing Replies always go back to WebChat CONFIRMED
Same session/routing as channels Unified session model CONFIRMED
chat.history Bounded conversation retrieval with truncation CONFIRMED
chat.send Send user messages CONFIRMED
chat.inject Append notes without triggering agent run CONFIRMED
Abort handling Partial output visible, abort metadata markers CONFIRMED
Read-only fallback When Gateway unreachable, history display only CONFIRMED
Tools panel Runtime tool catalog display in /agents section CONFIRMED
Static fallback tool list If dynamic catalog unavailable CONFIRMED

Status: CONFIRMED

5. Canvas / A2UI (Agent-Driven Visual Workspace)

Feature Description Status
Separate server process Port 18793, isolated from Gateway CONFIRMED
Real-time HTML rendering Agent pushes HTML over WebSocket CONFIRMED
A2UI attributes Agent-driven UI attributes embedded in HTML CONFIRMED
Charts and dashboards Agent-generated data visualizations CONFIRMED
Interactive interfaces User can interact with rendered content CONFIRMED
Live updates Content updates in real-time during agent runs CONFIRMED
Canvas push/reset Agent controls content lifecycle CONFIRMED
Canvas eval Execute JavaScript in canvas context CONFIRMED
Canvas snapshot Capture current canvas state CONFIRMED
Multi-platform rendering macOS (WebKit), iOS (SwiftUI), Android (WebView), Browser CONFIRMED
Crash isolation Canvas crash does not affect Gateway CONFIRMED

Status: CONFIRMED

6. Messaging Channels (24 Total)

Channel Library/Protocol Status
WhatsApp Baileys CONFIRMED
Telegram grammY CONFIRMED
Slack Bolt CONFIRMED
Discord discord.js CONFIRMED
Google Chat API CONFIRMED
Signal signal-cli CONFIRMED
iMessage BlueBubbles (recommended) CONFIRMED
iMessage Legacy Direct CONFIRMED
IRC - CONFIRMED
Microsoft Teams - CONFIRMED
Matrix - CONFIRMED
Feishu - CONFIRMED
LINE - CONFIRMED
Mattermost Plugin CONFIRMED
Nextcloud Talk - CONFIRMED
Nostr - CONFIRMED
Synology Chat - CONFIRMED
Tlon - CONFIRMED
Twitch - CONFIRMED
Zalo - CONFIRMED
Zalo Personal - CONFIRMED
WebChat Built-in CONFIRMED
macOS App Native CONFIRMED
iOS/Android Native CONFIRMED

Key channel features:

  • DM policies: pairing mode (default) or open mode
  • Channel allowlisting: per-channel allowFrom configuration
  • Group rules: mention gating, reply tags, per-channel chunking
  • Media pipeline: image, audio, video handling with transcription hooks and size caps
  • Typing indicators: presence and input signaling
  • Streaming/chunking: response fragmentation and reassembly

Status: CONFIRMED

7. Model Providers & Routing

7.1 Provider Support

Provider Integration Status
Anthropic (Claude) Native CONFIRMED
OpenAI (GPT) Native CONFIRMED
Google (Gemini) Native + OAuth CONFIRMED
OpenRouter Built-in (no config needed) CONFIRMED
Ollama Auto-detected at localhost:11434 CONFIRMED
AWS Bedrock Supported CONFIRMED
Azure OpenAI Supported CONFIRMED
Mistral Supported CONFIRMED
Voyage Supported (embeddings) CONFIRMED
Custom OpenAI-compatible Configurable base URL CONFIRMED

7.2 Model Routing Features

Feature Description Status
Failover chains Ordered fallback list across providers CONFIRMED
Auth profile rotation Multiple credentials per provider CONFIRMED
Exponential backoff When provider goes down CONFIRMED
Per-session model override Change model mid-session CONFIRMED
Per-cron model override Different model for scheduled tasks CONFIRMED
Model catalog Built-in list of supported models CONFIRMED
Custom model registration Add models via config CONFIRMED
Thinking/reasoning toggles Enable/disable extended thinking per model CONFIRMED

Status: CONFIRMED

8. Memory System

8.1 Core Storage

Feature Description Status
File-based memory Plain Markdown files as source of truth CONFIRMED
Daily logs memory/YYYY-MM-DD.md (append-only, auto-loaded) CONFIRMED
Long-term memory MEMORY.md (curated, private sessions only) CONFIRMED
Git-friendly Back up with Git, grep through, edit in any editor CONFIRMED
Workspace configurable Via agents.defaults.workspace CONFIRMED

8.2 Search & Retrieval

Feature Description Status
memory_search Semantic recall over indexed snippets CONFIRMED
memory_get Targeted file/line-range reading CONFIRMED
Hybrid search Vector similarity + BM25 keyword matching CONFIRMED
Score merging finalScore = vectorWeight * vectorScore + textWeight * textScore CONFIRMED
MMR re-ranking Diversity mode to reduce redundant snippets CONFIRMED
Temporal decay Exponential recency boost (configurable half-life, default 30 days) CONFIRMED
Evergreen files MEMORY.md exempt from temporal decay CONFIRMED
Extra paths Index additional directories via memorySearch.extraPaths CONFIRMED

8.3 Embedding Providers (auto-selected in order)

  1. Local GGUF models via node-llama-cpp
  2. OpenAI API
  3. Gemini embeddings
  4. Voyage embeddings
  5. Mistral embeddings
  6. Ollama (requires explicit config)

8.4 Advanced Memory Backends

Backend Description Status
Cognee Knowledge graph from conversational data, GRAPH_COMPLETION search CONFIRMED
Mem0 Auto-extracts structured facts, vector DB storage, deduplication CONFIRMED
Graphiti Temporal knowledge graph (Episodes, Entities, Communities, 4 timestamps) CONFIRMED
QMD Experimental sidecar combining BM25 + vectors + reranking CONFIRMED

8.5 Session Memory (Experimental)

Feature Description Status
Transcript indexing Opt-in via sessionMemory: true CONFIRMED
Delta-based triggers Async indexing with configurable thresholds CONFIRMED
Isolated per agent Never blocks memory_search calls CONFIRMED
QMD session export Sanitized session transcripts indexed for recall CONFIRMED

8.6 Memory Management

Feature Description Status
Pre-compaction flush Silent agentic turn before token limit CONFIRMED
Soft threshold Configurable softThresholdTokens (default 4000) CONFIRMED
File watcher Auto-reindex on file changes (debounce 1.5s) CONFIRMED
Embedding cache SQLite-backed to avoid re-embedding CONFIRMED
sqlite-vec acceleration Optional vector virtual table for fast queries CONFIRMED
Batch indexing Async bulk embedding with configurable concurrency CONFIRMED

Status: CONFIRMED

9. Skills / Plugin System (ClawHub)

9.1 Skill Structure

Feature Description Status
SKILL.md file Main skill definition in Markdown CONFIRMED
Supporting files Additional text/config files in folder CONFIRMED
Versioning Semver, changelogs, tags CONFIRMED
HOT/COLD split HOT section for quick reference, COLD for full docs CONFIRMED

9.2 ClawHub Registry

Feature Description Status
Public browsing Browse skills and SKILL.md content CONFIRMED
Vector search Embedding-powered search (not just keywords) CONFIRMED
Versioning Semver with changelog history CONFIRMED
Downloads Zip per version CONFIRMED
Stars and comments Community feedback CONFIRMED
Moderation hooks Approval and audit workflows CONFIRMED
CLI-friendly API Automation and scripting support CONFIRMED
13,729+ skills As of Feb 28, 2026 CONFIRMED
GitHub account gating Account must be 1+ week old to publish CONFIRMED

9.3 Skill Management in UI

Feature Description Status
Auto-discovery Agent can search ClawHub and install skills automatically CONFIRMED
Enable/disable per skill Toggle in Control UI CONFIRMED
API key per skill Manage credentials per skill CONFIRMED
Install gating Controlled skill installation CONFIRMED
Three levels Bundled, managed, workspace-level CONFIRMED

Status: CONFIRMED

10. Session Management

Feature Description Status
Session store Map of sessionKey -> { sessionId, updatedAt, ... } CONFIRMED
JSONL transcripts Per-session at ~/.openclaw/agents/<agentId>/sessions/ CONFIRMED
/status command Model, tokens, thinking/verbose state, credential freshness CONFIRMED
/compact command Summarize older context, persistent compaction CONFIRMED
/new command Start new session CONFIRMED
/reset command Reset session state CONFIRMED
Token tracking inputTokens, outputTokens, totalTokens, contextTokens CONFIRMED
Context pruning Automatic context management at limits CONFIRMED
Per-session thinking override Enable/disable reasoning per session CONFIRMED
Per-session verbose override Debug output toggle per session CONFIRMED
Agent bindings Route channels/users to specific agents CONFIRMED
Main/group isolation Separate DM and group session handling CONFIRMED
Activation modes Configurable trigger conditions CONFIRMED
Queue modes Message queuing behavior CONFIRMED

Status: CONFIRMED

11. Cron Jobs & Scheduling

Feature Description Status
Cron syntax Standard cron expressions CONFIRMED
Delay/one-shot "Poke agent in 20 minutes" type jobs CONFIRMED
Persistent jobs Survive gateway restarts CONFIRMED
Delivery to chat Optionally post results to a session CONFIRMED
Announce summary Default delivery mode CONFIRMED
Webhook delivery HTTP callback with optional bearer token CONFIRMED
Silent execution delivery: none for internal-only runs CONFIRMED
Delete-after-run One-shot execution CONFIRMED
Model overrides Per-job model selection CONFIRMED
Stagger option Prevent simultaneous execution CONFIRMED
Run history Track past executions CONFIRMED
Gmail Pub/Sub Integration for email-triggered jobs CONFIRMED

Status: CONFIRMED

12. Browser Automation

Feature Description Status
Dedicated Chrome/Chromium Separate browser instance CONFIRMED
CDP control Chrome DevTools Protocol integration CONFIRMED
Page snapshots Capture page state CONFIRMED
Actions Click, type, navigate, etc. CONFIRMED
File uploads Upload files to web pages CONFIRMED
Browser profiles Multiple profile support CONFIRMED
Accessibility tree Agent understands pages via a11y tree CONFIRMED
OAuth flows Automated authentication CONFIRMED
Web scraping Content extraction CONFIRMED

Status: CONFIRMED

13. Exec Approvals & Permissions

Feature Description Status
Exec tool Run shell commands on gateway or node host CONFIRMED
Allowlist mode Only pre-approved commands execute CONFIRMED
Per-agent allowlists Prevent cross-agent permission leakage CONFIRMED
Policy + allowlist + approval Three-layer permission check CONFIRMED
Shell chaining rules &&, `
Redirect restrictions Unsupported in allowlist mode CONFIRMED
Elevated toggle `/elevated on off` for bash access
Tool policy Allow/deny rules per agent and provider/channel CONFIRMED
Control UI management Edit defaults, per-agent overrides, allowlists CONFIRMED
JSON config file ~/.openclaw/exec-approvals.json CONFIRMED

Status: CONFIRMED

14. Nodes (Companion Devices)

14.1 Node Architecture

Feature Description Status
WebSocket connection role: "node" to Gateway CONFIRMED
Device pairing Identity presentation + pairing request CONFIRMED
Bonjour discovery iOS device pairing CONFIRMED
Command surface canvas.*, camera.*, device.*, notifications.*, system.* CONFIRMED
node.invoke Gateway dispatches commands to nodes CONFIRMED

14.2 Node Capabilities

Capability Description Status
Camera snap Take photos CONFIRMED
Camera clip Record video clips CONFIRMED
Screen record MP4 recording (max 60s) CONFIRMED
Location retrieval GPS coordinates CONFIRMED
Notifications Push notifications CONFIRMED
SMS access Read/send messages (Android) CONFIRMED
Photos access Photo library (Android) CONFIRMED
Contacts access Contact list (Android) CONFIRMED
Calendar access Calendar events (Android) CONFIRMED
Motion data Accelerometer/gyroscope (Android) CONFIRMED
App updates Application management (Android) CONFIRMED
System run Execute commands with --cwd, --env, --command-timeout CONFIRMED

14.3 Platform-Specific Nodes

Platform Features Status
macOS Menu bar app, Voice Wake, push-to-talk, WebChat, debug tools, SSH control CONFIRMED
iOS Canvas surface, Voice Wake, Talk Mode, camera, screen recording, Bonjour CONFIRMED
Android Connect/Chat/Voice tabs, Canvas, camera, screen capture, all device families CONFIRMED

Status: CONFIRMED

15. Voice & Audio

Feature Description Status
Talk Mode Continuous listen-think-speak loop CONFIRMED
Voice Wake macOS/iOS wake word detection CONFIRMED
Push-to-talk macOS manual voice activation CONFIRMED
Voice Activity Detection (VAD) Auto-detect speech start/end CONFIRMED
Interrupt-on-speech Stop playback when user speaks CONFIRMED
TTS providers ElevenLabs, Google Cloud TTS, Azure Speech, Coqui, system fallback CONFIRMED
STT providers Whisper API (OpenAI), Deepgram, Local Whisper (offline) CONFIRMED
Node mic loop Node handles microphone, Gateway handles model calls CONFIRMED
Audio transcription hooks Automatic voice-to-text on incoming audio CONFIRMED

Status: CONFIRMED

16. Cost / Budget Tracking

Feature Description Status
session_status tool Returns tokens in/out per run, model used CONFIRMED
Token counting Input/output/total/context tokens tracked CONFIRMED
Cost estimation Per-response cost when model provides data CONFIRMED
/usage command View token and cost data CONFIRMED
Model distribution tracking Which models are used how often CONFIRMED
NO native hard spend cap No built-in budget limit that cuts API calls CONFIRMED
Proxy workaround LiteLLM virtual keys with budget limits CONFIRMED (community)
Cron-based budget alert Custom skill to aggregate and alert CONFIRMED (community)
Dashboard cost cards Today's cost, all-time, projected monthly (community dashboards) CONFIRMED (community)

NOTABLE GAP: OpenClaw has NO built-in hard budget cap. This is a known pain point.

Status: CONFIRMED

17. Security & Secrets Management

17.1 Authentication

Feature Description Status
Gateway token auth WebSocket handshake authentication CONFIRMED
Password auth In-memory only CONFIRMED
Device pairing Browser device approval workflow CONFIRMED
Tailscale integration Tailnet-only HTTPS, identity verification CONFIRMED
OAuth flows Per-provider authentication CONFIRMED
SecretRef Managed token support CONFIRMED

17.2 Secrets Management

Feature Description Status
openclaw secrets CLI External secrets management tool CONFIRMED
HashiCorp Vault Integration via exec provider CONFIRMED
AWS Secrets Manager Integration via exec provider CONFIRMED
Runtime injection ${VAR} references resolved at startup CONFIRMED
No filesystem touch Secrets never written to disk CONFIRMED
Atomic reload Secrets refreshed on Gateway restart CONFIRMED
Exec provider bridge Generic adapter to any external secret manager CLI CONFIRMED

17.3 Security Hardening

Feature Description Status
openclaw security audit Built-in security audit command CONFIRMED
openclaw doctor Misconfiguration detection and remediation CONFIRMED
Allowlist mode Command execution restrictions CONFIRMED
Per-agent isolation Agent permissions don't leak to others CONFIRMED
TCC permission tracking macOS permission awareness CONFIRMED
Composio integration Managed auth, least-privilege, audit logs, kill switch CONFIRMED

Status: CONFIRMED

18. MCP (Model Context Protocol) Integration

Feature Description Status
Native MCP server support @modelcontextprotocol/sdk@1.25.3 CONFIRMED
Config-based server registration Specify in openclaw.json CONFIRMED
Tool exposure MCP server tools available to all agents CONFIRMED
1,000+ MCP servers Community-built servers available CONFIRMED
McPorter Discovery/install/management tool for MCP servers CONFIRMED
Bridge to Claude MCP server connecting OpenClaw to Claude.ai with OAuth2 CONFIRMED

Status: CONFIRMED

19. Desktop Apps

19.1 Official/Native Apps

Platform Features Status
macOS Menu bar app, TCC prompts, Voice Wake, push-to-talk, WebChat, debug tools, remote SSH CONFIRMED
iOS Canvas, Voice Wake, Talk Mode, camera, screen recording, Bonjour pairing CONFIRMED
Android Connect/Chat/Voice tabs, Canvas, camera, screen capture, device command families CONFIRMED

19.2 Community Desktop Apps

App Tech Stack Features Status
OpenClaw Desktop Tauri + SvelteKit + Ollama Chat, documents, games, NPC mode, multi-gateway CONFIRMED
OpenClaw-Windows - WSL/Native/Remote/WebSocket gateway connections CONFIRMED
EasyClaw Native Mac/Windows Simplified OpenClaw interface CONFIRMED
Claw Desktop Native Sessions, action approval, artifact review, proof export CONFIRMED
OpenClaw-Windows-Hub - System tray, shared library, node, PowerToys integration CONFIRMED

Status: CONFIRMED

20. Agent-to-Agent Communication

Feature Description Status
sessions_list Discover active agents and metadata CONFIRMED
sessions_history Fetch other agent transcripts CONFIRMED
sessions_send Message other sessions with reply-back and announce toggles CONFIRMED
Agent bindings Route channels/users to specific agents CONFIRMED
Thread-bound agents First-class runtime feature (2026.2.26+) CONFIRMED

Status: CONFIRMED

21. Chat Commands

Command Function Status
/status Model + tokens + toggles + credential status CONFIRMED
/new Start new session CONFIRMED
/reset Reset session CONFIRMED
/compact Summarize older context, free window space CONFIRMED
/think Toggle reasoning levels CONFIRMED
/verbose Toggle verbose output CONFIRMED
/usage Token and cost info CONFIRMED
/restart Restart gateway CONFIRMED
/activation Change activation mode CONFIRMED
/elevated on|off Toggle elevated bash access CONFIRMED
/stop Abort current run CONFIRMED

Status: CONFIRMED

22. Deployment & Installation

Method Description Status
openclaw onboard Interactive wizard (gateway, workspace, channels, skills) CONFIRMED
--install-daemon Install as launchd (macOS) / systemd (Linux) user service CONFIRMED
npm install npm install -g openclaw CONFIRMED
Docker Container-based deployment CONFIRMED
Nix Declarative configuration CONFIRMED
Stable/Beta/Dev channels npm dist-tags CONFIRMED
Tailscale Serve Tailnet-only HTTPS CONFIRMED
Tailscale Funnel Public HTTPS (password required) CONFIRMED
SSH tunnels Remote access with token/password CONFIRMED

Status: CONFIRMED

23. Localization

Feature Description Status
6 languages English, Simplified Chinese, Traditional Chinese, Portuguese (BR), German, Spanish CONFIRMED
Auto-detection Based on browser locale CONFIRMED
Language picker In Access card CONFIRMED
Lazy-loaded translations Non-English loaded on demand CONFIRMED
Fallback to English For missing translation keys CONFIRMED
Persistent locale Stored in browser storage CONFIRMED

Status: CONFIRMED

24. Known Weaknesses & Community Complaints

Critical Pain Points

Issue Severity Source
No hard budget cap HIGH Community consensus
Runaway API costs HIGH Reddit horror stories ($300-$750/mo)
Complex setup MEDIUM Server config, API management, troubleshooting
Security if misconfigured HIGH Shell access = potential backdoor
Name changes (3x) MEDIUM Documentation chaos with each rebrand
contextTokens reporting bug MEDIUM Shows model max instead of actual usage
Memory system "broken" MEDIUM Blog post: defaults not optimal for most users

Architecture Limitations

Limitation Impact
Node.js/TypeScript only No Python-native option
Gateway must stay running 24/7 Resource consumption
No built-in encryption at rest Files are plain Markdown
Memory relies on LLM quality Bad writes pollute memory
No native audit trail Relies on community tools for hash-chained audit
No built-in code execution sandbox Security concern for exec tool

Status: CONFIRMED from multiple community sources

25. Cato vs OpenClaw Gap Analysis

Features Cato ALREADY HAS that match/exceed OpenClaw

Feature Cato OpenClaw
Hard budget cap YES ($1 session / $20 monthly) NO (major Cato advantage)
Hash-chained audit log YES (SHA-256, tamper-proof) NO (community only)
AES-256-GCM vault YES (encrypted secrets) Partial (external vault integration)
Action guard (pre-execution gate) YES (3-rule system) Partial (exec approvals)
Reversibility registry YES (track reversible actions) NO
Delegation tokens YES (scoped, time-limited) NO
Knowledge graph memory YES (kg_nodes, kg_edges) Via plugins only (Cognee/Graphiti)
Multi-model orchestrator YES (Claude/Codex/Gemini/Cursor fan-out) YES (failover chains)
Query classification YES (TIER_A/B/C routing) NO (manual model selection)
Epistemic monitoring YES (premise extraction, gap detection) NO
Disagreement detection YES (multi-model Jaccard) NO
Decision memory YES (write_decision, record_outcome) NO
Contradiction detection YES (Jaccard 0.35 threshold) NO
Habit extraction YES (behavioral patterns) NO
Anomaly detection YES (domain monitoring) NO
Desktop app YES (Tauri v2 + React) Community only
Python-native YES NO (TypeScript only)

Features OpenClaw HAS that Cato LACKS

Feature Priority to Add Difficulty
24 messaging channels LOW (Cato is daemon-first) HIGH
Canvas / A2UI visual workspace MEDIUM MEDIUM
ClawHub skill marketplace (13K+ skills) LOW (Cato uses SKILL.md) HIGH
Voice / Talk Mode MEDIUM MEDIUM
Node system (companion devices) LOW HIGH
Cron scheduling MEDIUM (Cato has schedule_manager) LOW
Browser automation (CDP) MEDIUM (Cato uses patchright) LOW
MCP server integration HIGH MEDIUM
Localization (6 languages) LOW LOW
Config wizard (onboarding) MEDIUM LOW
File watcher for memory reindex LOW LOW
MMR re-ranking for search MEDIUM MEDIUM
Temporal decay for memory search MEDIUM MEDIUM
Session compaction (/compact) MEDIUM MEDIUM
Pre-compaction memory flush MEDIUM LOW
QMD experimental backend LOW HIGH

Cato's UNIQUE Advantages (OpenClaw Cannot Match)

  1. Hard budget enforcement — OpenClaw has zero native spend caps; Cato has $1/$20 hard limits
  2. Tamper-proof audit chain — SHA-256 hash-chained, field-level re-hash verification
  3. Action guard with 3-rule gate — Pre-execution safety checks before any tool runs
  4. Reversibility tracking — Know which actions can be undone
  5. Delegation tokens — Scoped, time-limited, spending-capped authorization
  6. Epistemic monitoring — Detect when the AI doesn't know enough to answer well
  7. Multi-model disagreement surfacing — Catch when different models disagree
  8. Decision memory with outcome tracking — Record decisions, track if they were correct
  9. Contradiction detection — Automatically find conflicting information in memory
  10. Python-native — Simpler deployment for Python shops, no Node.js dependency

Sources
⚡