Research Raw Data: OpenClaw Competitive Intelligence

Date: February 17, 2026 Purpose: Supporting data for the main Competitive Intelligence Report

1. OpenClaw Cost Horror Stories (Verified Reports)

Cost Breakdown by Category

• Context accumulation: 40–50% of total spend
• System prompt overhead: 10–15% (SOUL.md + AGENTS.md + skills = 5K–10K tokens/call)
• Cron/heartbeat: Each trigger = new conversation = full context re-injection
• 24/7 operation with API models: $800–1,500/month

Cost Optimization Strategies (Community-Sourced)

• ClawRouter: Auto-routes to cheapest capable model, saves ~90%
• Prompt caching: Cuts bills 50–70% (same system prompt only charged full once)
• Local models via Ollama: Qwen3 8B, Llama 3.2 (requires 16GB+ RAM, zero API cost)
• Model tiering: Haiku/Flash for simple tasks, Sonnet/Opus for complex only
• Budget caps + spending alerts

2. Security Audit Findings

Bitdefender Audit

• 135,000+ OpenClaw instances exposed publicly on the internet
• Default bind to 0.0.0.0 (every network interface)
• Many vulnerable to remote code execution

Snyk Scan of ClawHub

• Scanned entire marketplace: ~4,000 skills
• 283 skills (7.1%) expose sensitive credentials
• API keys stored in plaintext

Koi Research Audit of ClawHub

• 341 malicious skills out of 2,857 total (~12%)
• 335 attempted to infect Apple computers with Atomic Stealer malware
• Skills used professional documentation and innocuous names (e.g., "solana-wallet-tracker")
• Keyloggers and credential stealers found

CVE-2026-25253

• One-click remote code execution vulnerability
• Exploited Control UI's trust of URL parameters without validation
• Patched in version 2026.1.29 before public disclosure

Config Redaction Bug (Issue #11268)

• Running openclaw configure in v2026.2.6-3 writes __OPENCLAW_REDACTED__ placeholder to actual config file
• Destroys all API keys, tokens, and numeric configuration values

Expert Quotes

• Cisco: "An absolute nightmare" from a security perspective
• Northeastern Prof. Aanjhan Ranganathan: "A privacy nightmare"
• Security researcher Ian Ahl: "AI agents get credentials to your entire digital life. Unlike browser extensions that run in a sandbox, these agents operate with the full privileges you grant them."
• OpenClaw maintainer "Shadow" on Discord: "If you can't understand how to run a command line, this is far too dangerous for you to use safely"

3. GitHub Repository Stats & Top Issues

Repository Stats (as of Feb 17, 2026)

• Stars: 197,171
• Forks: 34,310
• Open Issues: 5,825+
• Language: TypeScript
• Codebase: 430,000+ lines
• Contributors: 488

Top Issues by Comments

Recent Release Highlights

v2026.2.6: Anthropic Opus 4.6 support, xAI (Grok) provider, OpenAI Codex gpt-5.3-codex v2026.2.2: Feishu/Lark plugin, Web UI Agents dashboard, QMD memory backend, security approvals v2026.2.1: Memory search L2-normalize fix, Windows spawn() fix, Slack hardening

4. OpenClaw Alternatives Detailed Comparison

ZeroClaw (Rust-Based)

• Memory: <5MB idle (vs OpenClaw's 394MB idle, 1.52GB active)
• Startup: <10ms (400x faster)
• Binary size: 3.4MB single self-contained binary
• Storage: Hybrid SQLite (vector embeddings as BLOBs, FTS5 with BM25 scoring)
• Platforms: ARM, x86, RISC-V
• Trade-off: Extending requires Rust knowledge

NanoClaw (Security-Focused)

• Built as direct reaction to OpenClaw's security architecture
• Agent runs inside isolated Docker containers
• No access to host filesystem or environment variables outside designated directory
• "Even if the AI goes rogue, it can only mess up the sandbox"

Nanobot (Minimalist Python)

• 4,000 lines of Python (99% smaller than OpenClaw's 430,000 lines)
• University of Hong Kong project
• Core features: persistent memory via knowledge graph, web search, background sub-agents, Telegram + WhatsApp
• "Entire codebase can be understood in a few days instead of months"

memU (Memory-Focused)

• Builds local knowledge graph of preferences, past projects, habits
• "Not god-mode — smart-mode"
• Focus on being the smartest assistant, not the most capable
• Lower API costs through efficient context management

Moltworker (Serverless)

• OpenClaw capabilities without running on personal machine
• Cloud-hosted approach

5. Competitor Pricing Summary

Dust.tt Details

• SOC 2 Type II certified, GDPR compliant, HIPAA-ready
• SSO: Okta, Entra, JumpCloud
• SCIM provisioning
• US/EU data residency
• Enterprise audit logs
• $16M Series A (Sequoia Capital)
• Multi-model: OpenAI, Anthropic, Gemini, Mistral

Devin Real-World Performance

• Test results: "From 20 tasks, Devin failed 14 times, succeeded 3 times, unclear 3 times"
• SWE-bench: 13.86% of real-life GitHub issues solved
• REST API test: 45 minutes, 10/15 endpoints correct
• ACU pricing model is unpredictable — budgeting is difficult
• Best for: migrations, bulk refactoring (repetitive, well-scoped tasks)

Cursor Performance

• 126% productivity boost reported by users
• 30–40% acceleration on routine coding tasks
• Philosophy: "You think through the code" (interactive, not autonomous)

Claude Code

• SWE-bench: 80.9% accuracy with Opus 4.5 (industry leader for accuracy)
• Terminal-based; no GUI

6. Expert & Media Quotes

Chris Symons, Chief AI Scientist at Lirio (TechCrunch):

"OpenClaw is just an iterative improvement on what people are already doing, and most of that iterative improvement has to do with giving it more access."

Artem Sorokin, AI Engineer:

"From an AI research perspective, this is nothing novel."

MacStories reviewer:

"OpenClaw showed me what the future of personal AI assistants looks like."

Community sentiment:

"Open source built a better version of Siri that Apple ($3.6 trillion company) was sleeping on for years."

IBM's Kaoutar El Maghraoui:

OpenClaw "challenges the hypothesis that autonomous AI agents must be vertically integrated, with the provider tightly controlling the models, memory, tools, interface, execution layer and security stack."

Peter Steinberger (Creator):

Predicts that OpenClaw-style agents "will kill 80% of apps." Announced joining OpenAI on Feb 14, 2026.

7. Community Ecosystem

• ClawHub: 5,700+ skills (marketplace)
• awesome-openclaw: 3,000+ curated resources on GitHub
• Moltbook: Social network exclusively for AI agents (1.6M registered bots, 7.5M AI-generated posts)
• Clawe: Open-source coordination layer for multiple OpenClaw agents
• ClawRouter: Third-party smart model routing (saves ~90% on costs)
• Derivative projects: NanoClaw, ZeroClaw, Nanobot, TapnClaw, EZClaw, TinyClaw (~400 lines shell script)