GoClaw

// Overview

GoClaw is OpenClaw's agent model rebuilt in Go and reframed as a gateway you can run for more than one person. Where the original is a single-user runtime, GoClaw is multi-tenant from the database up: per-user workspaces, per-user context files, AES-256-GCM-encrypted API keys, RBAC and isolated sessions on a PostgreSQL 18 schema with pgvector. The payoff of the rewrite is operational — a roughly 25 MB static binary with no Node.js runtime, sub-second startup, and a footprint small enough for a $5 VPS.

Every request runs an eight-stage pipeline — context, history, prompt, think, act, observe, memory, summarize — whose stages are pluggable but always executed, so behaviour is predictable rather than emergent. Prompt construction is its own subsystem: four modes (Full, Task, Minimal, None) with section gating, cache-boundary optimisation and per-session mode resolution, which is how a long-running agent keeps prompt-cache hits while still varying what it sends.

Memory is three tiers rather than one store: working memory for the live conversation, episodic memory for session summaries, and semantic memory as a knowledge graph, loaded progressively at L0/L1/L2. Alongside it sits a Knowledge Vault — a document registry with `[[wikilinks]]`, hybrid FTS plus pgvector search, and filesystem sync. Agents can be organised into teams with shared task boards, delegate to each other synchronously or asynchronously, and be orchestrated in automatic, explicit or manual modes. A self-evolution loop turns metrics into suggestions and, within guardrails, lets agents adapt their own communication style.

Reach is the other half of the pitch: 20+ LLM providers (Anthropic over native HTTP+SSE with prompt caching, OpenAI, OpenRouter, Groq, DeepSeek, Gemini, Mistral, xAI, MiniMax, DashScope, Claude CLI, Codex, ACP and any OpenAI-compatible endpoint) and seven messaging channels (Telegram, Discord, Slack, Zalo OA, Zalo Personal, Feishu/Lark, WhatsApp). Security is stated as five layers — a permission system, rate limiting, prompt-injection detection, SSRF protection and encryption. LLM call tracing with spans and prompt-cache metrics is built in, with optional OpenTelemetry OTLP export. A desktop edition, GoClaw Lite, drops Docker and PostgreSQL for a Wails + React app on SQLite, capped at five agents and one team, with FTS5 search in place of pgvector and no channels. Note the licence: GoClaw is distributed under Creative Commons Attribution-NonCommercial 4.0, not an open-source licence.

// What it does

• An 8-stage agent pipeline — context → history → prompt → think → act → observe → memory → summarize — with pluggable, always-on stages
• A 4-mode prompt system (Full / Task / Minimal / None) with section gating, cache-boundary optimisation and per-session mode resolution
• 3-tier memory — working, episodic, semantic knowledge graph — with progressive L0/L1/L2 loading, plus a Knowledge Vault with [[wikilinks]] and hybrid FTS + pgvector search
• Agent teams: shared task boards, synchronous and asynchronous inter-agent delegation, and automatic, explicit or manual orchestration
• Multi-tenant PostgreSQL with per-user workspaces and context files, AES-256-GCM-encrypted keys, RBAC and isolated sessions
• 20+ LLM providers and 7 messaging channels (Telegram, Discord, Slack, Zalo OA, Zalo Personal, Feishu/Lark, WhatsApp) behind one gateway
• Built-in LLM call tracing with spans and prompt-cache metrics, plus optional OpenTelemetry OTLP export and a `goclaw traces` operator CLI
• A ~25 MB single static binary with sub-second startup, and a Docker path with optional headless Chrome, Jaeger, sandbox, Tailscale and Redis services
• GoClaw Lite — a native Wails + React desktop app on SQLite with no Docker or PostgreSQL, capped at 5 agents and 1 team

// Getting started

GoClaw needs Go 1.26+ and PostgreSQL 18 with pgvector; Docker is optional but is the quicker route. Note that the repository's default branch is `dev` — clone `main` for the stable release branch.

git clone -b main https://github.com/nextlevelbuilder/goclaw.git && cd goclaw
make build
./goclaw onboard        # interactive setup wizard
source .env.local && ./goclaw

chmod +x prepare-env.sh && ./prepare-env.sh

# add at least one GOCLAW_*_API_KEY to .env, then:
make up

curl http://localhost:18790/health

make up WITH_BROWSER=1 WITH_OTEL=1
make down WITH_BROWSER=1 WITH_OTEL=1

goclaw traces list --status error
goclaw traces get <trace-id> -o json
goclaw --server https://goclaw.example.com --token "$GOCLAW_GATEWAY_TOKEN" traces follow --session <session-key>

# macOS
curl -fsSL https://raw.githubusercontent.com/nextlevelbuilder/goclaw/main/scripts/install-lite.sh | bash

# Windows (PowerShell)
irm https://raw.githubusercontent.com/nextlevelbuilder/goclaw/main/scripts/install-lite.ps1 | iex

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Reach for it when several people need their own agents on shared infrastructure, with credentials and sessions genuinely isolated
• Reach for it when an agent has to be reachable from chat — Telegram, Discord, Slack, Feishu, WhatsApp — rather than from a terminal
• Reach for it when the deployment target is small: one static binary and a Postgres instance instead of a Node runtime and a process tree
• Reach for it when a team of agents needs a shared task board and explicit delegation rather than one agent doing everything

// How GoClaw compares

GoClaw alongside other open-source agent frameworks & builders tools AI/TLDR tracks, ranked by GitHub stars.