DISCOVER THE FUTURE OF AI AGENTSarrow_forward

Golemancy

calendar_todayAdded Apr 25, 2026
categoryAgent & Tooling
codeOpen Source
Workflow AutomationElectron桌面应用大语言模型Multi-Agent SystemModel Context ProtocolAI AgentsAgent FrameworkBrowser AutomationAgent & ToolingModel & Inference FrameworkAutomation, Workflow & RPAProtocol, API & Integration

Free open-source desktop AI team orchestration tool featuring multi-agent, recursive sub-agents, visual topology editing, browser automation, and MCP protocol support, empowering individuals with a full AI team's collaborative capabilities.

Golemancy is a free open-source desktop application developed by Jicai, Inc., positioned as a personal AI team orchestration platform. Its core mechanism revolves around multi-agent architecture: each agent has independent context, tools, and tasks, with support for recursive sub-agent nesting to trigger complete autonomous workflows. Through a visual topology editor, users can organize multiple agents into structured teams, defining leader-member hierarchies and running entire teams within a single conversation.

For model integration, Golemancy supports 9+ LLM providers including Claude, GPT, Gemini, and DeepSeek, with compatibility for any OpenAI-format API, allowing per-agent model switching. The tool layer natively integrates the MCP protocol (with connection pooling) for direct access to the MCP tool ecosystem, and includes built-in Playwright-based browser automation with 16 tools and 80+ operations. The skill system operates as reusable prompt templates, supporting creation, sharing, and import. Agent memory uses per-agent isolated SQLite persistent storage with tiered management of fixed facts and time-sensitive memories. Additional capabilities include voice input, Cron scheduling, and a built-in file workspace.

Architecturally, it adopts a Monorepo structure where the Electron shell forks a Hono HTTP Server child process. All communication runs over localhost loopback, with per-project independent SQLite databases enforcing a local-first data security strategy. The frontend is built on React + Zustand, the backend on Hono + better-sqlite3 + drizzle-orm, and the AI invocation layer uniformly uses Vercel AI SDK. Current version is v0.1.73, licensed under Apache License 2.0.

Use Cases

  • Research assistant: parallel multi-agent information gathering and organization
  • Content creation: writing, translation, marketing copy generation
  • Data analysis: automated data processing and report generation
  • Web scraping and monitoring: periodic data collection via browser automation
  • Personal AI team: one person as a team, lowering multi-role collaboration barriers

Security Mechanisms

  • Local-first: data never leaves the machine, loopback communication only
  • Session authentication: per-session Bearer Token
  • Three-level sandbox: permission control (specific tier details pending confirmation)

Unconfirmed Information

  • Discord and Twitter links exist as anchors in README but actual URLs are not visible
  • MCP connection pool configuration, max connections, and timeout policies are not detailed
  • Recursive sub-agent nesting lacks actual performance benchmarks (latency, memory, concurrency)
  • Three-level sandbox permission tier definitions and granularity are not publicly disclosed
  • Whether Cron scheduled tasks continue after desktop app closure is not specified

Related Projects

View All arrow_forward

STAY UPDATED

Get the latest AI tools and trends delivered straight to your inbox. No spam, just intelligence.

rocket_launch