GoGogot

GoGogot is an open-source, self-hosted AI agent written in Go that ships as a single ~15 MB binary idling at ~10 MB RAM. It supports 7 built-in model aliases (Claude, GPT, DeepSeek, Gemini, MiniMax, Qwen, Llama, Kimi) and any OpenRouter slug, making it a lightweight alternative to heavier agent frameworks. Deployable with a single Docker command, it gives users full control over API keys, model selection, and data — nothing phones home. With only 8 dependencies and ~4.5K lines of Go, it is auditable, forkable, and MIT licensed.

• 27 Built-in Tools across 10 categories including Bash/shell, file read/write/edit, web search, HTTP requests, downloads, Telegram multi-chat, cron scheduling, memory, skills, tasks, and observability — all shipped in the binary.
• Multi-Model Support — set GOGOGOT_PROVIDER and GOGOGOT_MODEL to switch between Anthropic, OpenAI, or any OpenRouter slug without changing code.
• Persistent Identity & Memory — agent maintains soul.md and user.md for persistent identity, plus markdown-based memory files it manages autonomously.
• Reusable Skills — define procedural knowledge once and reuse it across sessions for repeatable automation workflows.
• Cron Scheduling — schedule tasks with cron expressions; schedules persist across restarts.
• Auto Context Compaction — automatically compresses context near token limits to keep long sessions running smoothly.
• Pluggable Transports — Telegram, HTTP, or bring your own transport by implementing a single interface.
• Structured Observability — emits structured events for LLM calls, tool invocations, and errors for easy debugging and auditing.
• One-Command Deploy — run docker run with your API keys and Telegram bot token to go live instantly on any Linux server or cheap VPS.
• MIT Licensed & Open Source — full source on GitHub; read, fork, and audit every line.