StackMap
Subscribe

mateclaw vs picobot

Self-hosted Java agent platform: ReAct plus plan-and-execute 'digital employees' with an LLM Wiki knowledge layer, skills/MCP/ACP tools, approval-gated actions and eight IM channels in one JAR. — versus — A self-hosted personal AI agent in a single ~9MB Go binary — persistent memory, 16 tools + MCP, skills, cron/heartbeat, and Telegram/Discord/Slack/WhatsApp channels. Runs on a $5 VPS.

The curated verdict

Same shape — personal agent on your own box with memory, skills, MCP and chat channels — at opposite scales: picobot is a 9MB Go binary on a $5 VPS, MateClaw a Spring Boot deployment with multi-user workspaces and consoles.

mateclawpicobot
Stars1.0k1.3k
Forks318166
LanguageJavaGo
LicenseApache-2.0MIT
Last activitytoday4 months ago
Topicsagents, orchestration, memoryagents
Curated connections53

mateclaw — the curator's take

Pick it when the buyer is an IT department rather than a hacker: multi-user workspaces, RBAC and approval gates on sensitive tool calls, audit trail, Spring Actuator health, per-channel error isolation, and provider failover across DashScope/OpenAI/Anthropic/Gemini/DeepSeek/Ollama when a vendor 401s mid-request. Underneath it's a real harness — StateGraph runtime, Team Runs with a durable runId, task DAGs, human approval — not a RAG call in a wrapper. The cost is weight and gravity: Java 21 + Spring Boot 3.5 + Vue, an enormous and fast-moving feature surface (workflows, triggers, Wiki transformations, Dream consolidation, skill mining), and a China-first channel set (DingTalk, Feishu, WeChat Work, QQ) whose value drops sharply outside that ecosystem. For a personal agent on a $5 VPS this is orders of magnitude too much machinery.

picobot — the curator's take

The anti-bloat statement piece: zero dependencies, ~10MB RAM idle, instant cold start — an always-on agent with ranked memory recall, background subagents, a natural-language HEARTBEAT.md cron, and self-authored skills ('create a skill for checking weather' → it writes the markdown), reachable from your phone via four chat channels. Runs on a Raspberry Pi or Termux on an old Android. Any OpenAI-compatible endpoint, including Ollama. Its own README names the target: OpenClaw's power without the 500MB container. NOT a coding harness and single-user by design — this is your personal daemon, not a team platform; complex multi-agent workflows will outgrow it fast, which is rather the point.