Introduction to GoClaw
Overview
GoClaw is an AI agent gateway written in Go — a multi-tenant platform for deploying and orchestrating LLM-powered agents across multiple messaging channels and protocols.
GoClaw is a Go port of OpenClaw, completely rewritten with:
- Enhanced security (5 defense layers, AES-256-GCM)
- Multi-tenant PostgreSQL with per-user workspace isolation
- Single binary ~25 MB, starts in under 1 second, runs on a $5 VPS
The gateway receives messages from channels (Telegram, Discord, etc.), routes them through the Agent Loop, calls LLM providers, executes tools, and returns results to the user.
Key Features
- 7 channels — Telegram, Discord, Slack, Feishu/Lark, Zalo OA, Zalo Personal, WhatsApp
- 20+ LLM providers — Anthropic (native HTTP+SSE), OpenAI, OpenRouter, Gemini, DeepSeek, Groq, Mistral, xAI, DashScope, Ollama, and more
- Agent teams & orchestration — Shared task board, inter-agent delegation (sync/async), team mailbox
- Tools & skills — Filesystem, exec/shell, web search, memory, browser automation, TTS, media (image/audio/video generation)
- MCP — Model Context Protocol bridge (stdio / SSE / streamable-HTTP)
- Scheduling & cron —
at,every, cron expression with lane-based concurrency - Security — Rate limiting, prompt injection detection, SSRF protection, RBAC (admin/operator/viewer), path traversal prevention
- Tracing & observability — LLM call tracing with spans, prompt cache metrics, optional OpenTelemetry OTLP export
- Extended thinking — Anthropic budget tokens, OpenAI reasoning effort, DashScope thinking budget
- Knowledge graph — LLM extraction + traversal
- Multi-tenant — Per-user workspace, encrypted API keys, isolated sessions
Architecture Overview
+----------------------------------------------------------+
| CHANNELS |
| Telegram Discord Slack Feishu Zalo OA Zalo WhatsApp|
+------------------------+---------------------------------+
|
+------------------------v---------------------------------+
| GATEWAY |
| WebSocket RPC v3 + HTTP API |
| Rate Limiter -> Permission Engine (RBAC) |
+------------------------+---------------------------------+
|
+------------------------v---------------------------------+
| AGENT LOOP |
| Think -> Act -> Observe |
| Scheduler (4 lanes: main / subagent / team / cron) |
+-------------+---------------------------+----------------+
| |
+-------------v-----------+ +-----------v---------------+
| LLM PROVIDERS | | TOOL REGISTRY |
| Anthropic OpenAI ... | | fs exec web memory |
| 20+ providers | | teams browser TTS |
+-------------------------+ +---------------------------+
|
+------------------------v---------------------------------+
| STORE LAYER |
| PostgreSQL (or SQLite for Lite) |
| Sessions Agents Memory Skills Traces Teams MCP |
+----------------------------------------------------------+Two Editions
| Feature | Lite (Desktop) | Standard (Server) |
|---|---|---|
| Database | SQLite (local) | PostgreSQL |
| Agents | Up to 5 | Unlimited |
| Teams | 1 team, 5 members | Unlimited |
| Memory | FTS5 text search | pgvector semantic |
| Channels | Not supported | Telegram, Discord, Slack, Zalo, Feishu, WhatsApp |
| Knowledge Graph | Not supported | Full |
| RBAC / Multi-tenant | Not supported | Full |
| Installation | One-line script (macOS/Windows) | Docker or binary |
| Auto-update | GitHub Releases | Docker / binary |
Lite is suitable for personal use on a desktop computer, with no need for a separate server or database.
Standard is designed for production deployments: serving multiple users, multiple tenants, and connecting to real messaging channels.
System Requirements
Standard (Server)
| Component | Requirement |
|---|---|
| Go | 1.26+ |
| PostgreSQL | 18+ with pgvector extension |
| Docker | Optional — used for sandbox and quick-start |
| Node.js | Optional — required if using MCP server via stdio |
Minimum RAM: ~35 MB (idle). Runs on a $5 VPS.
Lite (Desktop)
| Component | Requirement |
|---|---|
| macOS | Apple Silicon or Intel |
| Windows | 64-bit |
| Disk space | ~30 MB |
No Go, PostgreSQL, or Docker required.