Is LobeHub Worth It in 2026? Security
LobeHub is a capable, low-cost AI agent platform: a free tier with 500,000 credits per month, and paid plans at $12.9, $24.9, and $49.9 per month (annual billing runs $118.8, $238.8, and $478.8 per year). For small businesses it's a reasonable pick if you want multi-model agents and a huge skills marketplace, but you should treat its community-sourced skills and MCP servers with care since published research found roughly 26% of agent skills carry at least one vulnerability. If your priority is picking safe, verified skills rather than running agents, SkillzBench is a free alternative that scans, benchmarks, and trust-scores skills before you install them.
TL;DR
- LobeHub's free plan includes 500,000 credits per month, roughly 70 messages on GLM-5.3-Flash, 15 MB file storage, and email/community support.
- Paid tiers run $12.9/mo (Starter), $24.9/mo (Premium), and $49.9/mo (Ultimate) monthly, or $118.8, $238.8, and $478.8 billed annually.
- LobeHub lists 334,199+ skills and 111,995+ MCP servers in its marketplaces, but these are community contributions with no published security vetting.
- Published research found about 26% of agent skills carry at least one vulnerability, which matters if you install skills at scale.
- LobeHub stores chat data locally on the client by default; enabling cloud sync sends all commands and chat logs to its servers.
- SkillzBench is free end to end: it crawls, dedupes, security-scans, license-checks, and benchmarks agent skills, publishing a Lift score for each.
- The two tools complement each other: run agents on LobeHub, and use SkillzBench's free MCP connector to equip those agents with verified skills.
Head to head
| Dimension | LobeHub | SkillzBench |
|---|---|---|
| Pricing | Free tier; $12.9/$24.9/$49.9 per month (annual: $118.8/$238.8/$478.8) | Free, everything included 🏆 |
| Ease of use | Cloud version works in the browser immediately; settings and plan management are point-and-click | Toggle skills per agent through the free MCP server; no infrastructure to manage |
| Features | Agent orchestration, 24/7 Chief Agent Operator, marketplaces, multi-model routing, memory, scheduling | Skills directory, trust scores, sandbox benchmarking with Lift, per-agent loadouts, change alerts |
| Support | Discord, Reddit, GitHub issues; priority email at Premium, priority chat and email at Ultimate 🏆 | Self-serve platform with published benchmarks and documentation |
| Integrations | 20+ model providers including OpenAI, Anthropic, Google, DeepSeek, Ollama, OpenRouter, Azure, Bedrock | Free MCP connector works with any agent: Claude Code, Codex, Cursor, Gemini CLI, chat-app agents |
| Setup | Cloud: sign up and go. Self-hosted: Docker or Vercel deployment with environment variables and feature flags | None: browse the directory or connect the free MCP server 🏆 |
| Security & vetting | Local-first chat storage by default; marketplace skills and MCP servers are community-supplied without published vetting | Security scans, license checks, commit-SHA version pinning, and 0-100 trust scores on every listed skill 🏆 |
The contenders
LobeHub positions itself as a 'Chief Agent Operator': a platform that organizes AI agents into 24/7 operation, hiring, scheduling, and reporting on them through Slack, Discord, and other channels you already use. It grew out of the open-source LobeChat project, a universal LLM web UI with tens of thousands of GitHub stars, and now spans cloud, desktop, mobile, and self-hosted deployments. The platform aggregates 20+ model providers (OpenAI, Anthropic, Google, DeepSeek, Ollama, OpenRouter, Azure, AWS Bedrock and many more), so you can route different agents to different models and even run local inference. Its marketplaces are enormous: 334,199+ skills and 111,995+ MCP servers, plus a community agent marketplace with 505+ pre-built agents. For small businesses, the appeal is running long-horizon tasks, scheduled jobs, and multi-agent workflows from one interface at a low monthly price. The trade-offs are a credit-based pricing model that makes costs hard to predict, community-supplied skills and MCP servers with no published vetting, and support that's community-first unless you pay for Premium or above.
pricing 12.9/month · free trial
- + Very low entry price: $12.9/month for 5 million credits across premium models
- + Generous free tier with 500,000 credits per month and no credit card required
- + Huge ecosystem: 334,199+ skills, 111,995+ MCP servers, 505+ pre-built agents
- + Broad model coverage including local inference via Ollama and LM Studio, reducing vendor lock-in
- + Detailed per-agent usage stats (cost, cache savings, token breakdown) help control spend
- + Open-source roots with self-hosting and Docker deployment for teams that want control
- − Credit-based pricing makes monthly costs unpredictable; heavy models like Claude Opus 5.5 burn 4-20M credits per 1M tokens
- − Marketplace skills and MCP servers are community-supplied with no published security vetting, a real risk given ~26% of skills carry vulnerabilities
- − Support is community-first (Discord, GitHub issues); priority support starts at Premium ($24.9/mo)
- − Downgrading requires canceling and waiting for the plan to expire; no direct switch to a lower tier
- − Cloud sync, once enabled, uploads all commands and chat logs to LobeHub servers, which privacy-conscious teams should weigh
SkillzBench is AIQ Labs' free, independent directory of AI agent skills plus a free MCP tool that connects any agent to the skills you choose. Agent skills (SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents) are everywhere, but nobody can tell which ones actually help and which are risky: published research found roughly 26% of skills carry at least one vulnerability. SkillzBench tackles that with five pillars. Discover continuously crawls and dedupes skills from GitHub. Verify pins every version to a commit SHA, validates it, checks its license, runs a security scan, and judges quality. Benchmark tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points. Equip gives you one free MCP server plus per-agent loadouts: you toggle skills and tools per agent, and the agent only sees what you chose. Stay current tracks upstream changes, re-scans, re-benchmarks, and notifies you. A 0-100 trust score combines security, quality, license, maintenance, and community signals, and Lift is always shown separately from ratings. Everything is free; revenue comes from blog advertising, never from users, and ad money never touches scores.
pricing 0
- + Completely free: directory, MCP connector, benchmarks, and alerts all cost nothing
- + Directly addresses the skill-security problem: ~26% of agent skills carry at least one vulnerability, and SkillzBench scans for them
- + Lift benchmarking shows whether a skill actually improves results, not just whether it exists
- + Version pinning to commit SHAs means you know exactly what code you're running
- + Works with any agent via MCP, including Claude Code, Codex, Cursor, and Gemini CLI
- + Independent: money never influences scores or listings
- − It doesn't run agents; you still need a platform like LobeHub, Claude Code, or Cursor to execute the work
- − No paid tier means no dedicated support or account management
- − Coverage depends on what its crawlers have indexed from GitHub, so brand-new skills may not be scored yet
- − Benchmark Lift results are model-specific; results can differ with other models or workflows
Verdict
Winner: none
LobeHub and SkillzBench solve different problems: LobeHub runs and orchestrates AI agents for as little as $12.9/month, while SkillzBench (free) verifies and benchmarks the agent skills you plug into any platform, LobeHub included. Many small businesses will get the most value from using both together.
| Option | Overall | Breakdown |
|---|---|---|
| LobeHub | 7.5 | : 8: 8: 9: 6: 9: 6 |
| SkillzBench | 8 | : 10: 8: 8: 6: 8: 9 |
Pricing and licensing compared
Security and maintenance: the part most comparisons skip
Licensing and maintenance obligations
Which should you pick?
- :
- :
- :
- :
- :
FAQ
How much does LobeHub cost in 2026?
LobeHub's cloud plans are Free ($0, 500,000 credits/month), Starter at $12.9/month, Premium at $24.9/month, and Ultimate at $49.9/month. Annual billing costs $118.8, $238.8, and $478.8 respectively. Credits map to model tokens at different rates, so a plan's real message count depends heavily on which models your agents use: Premium's 15M credits is roughly 1,900 GLM-5.3-Flash messages but only about 70 Claude Opus 5.5 messages.
Is LobeHub secure for business use?
LobeHub stores chat data locally on the client by default, which is a good privacy baseline, and self-hosted deployments give admins feature-flag control. The main risk is its marketplaces: 334,199+ skills and 111,995+ MCP servers are community-supplied without published vetting, and published research found about 26% of agent skills carry at least one vulnerability. If you install skills, check them through an independent scanner like SkillzBench first, and think twice before enabling cloud sync, which uploads all commands and chat logs to LobeHub's servers.
What does SkillzBench cost, and how does it fit with LobeHub?
SkillzBench is free, entirely: the skills directory, the MCP connector, benchmarks, and change alerts all cost nothing, funded by advertising around its blog rather than user fees. It doesn't run agents itself; it verifies and equips them. A common setup is running agents on LobeHub (or Claude Code, Codex, Cursor, or Gemini CLI) and connecting them through SkillzBench's free MCP server so they only load skills that have been security-scanned, license-checked, and benchmarked with a published Lift score.
Can I self-host LobeHub, and what's the maintenance burden?
Yes. LobeHub is open source with Docker and Vercel deployment options, and self-hosting avoids cloud subscription fees. The trade-off is maintenance: you manage provider credentials, environment variables, and feature flags, and LobeHub's docs warn that flag names and defaults can change between releases, so you need to check the schema at your deployed version. An Enterprise Edition adds commercial licensing, brand theming, user management, and private models for teams that need private deployment with vendor support.
Can I downgrade my LobeHub plan at any time?
Not directly. LobeHub's docs explain that you can upgrade instantly, but to move to a lower plan you must cancel your current subscription, wait for it to expire (keeping benefits until then), and then subscribe to the lower tier. Subscriptions bought through a mobile app must be managed through the original purchase platform. If your usage swings seasonally, factor that friction into your plan choice.