// the find
nextlevelbuilder/goclaw
GoClaw - GoClaw is OpenClaw rebuilt in Go — with multi-tenant isolation, 5-layer security, and native concurrency. Deploy AI agent teams at scale without compromising on safety.
GoClaw is a Go-based multi-tenant AI agent gateway — a single binary that handles LLM routing across 20+ providers, agent orchestration, 3-tier memory (working/episodic/semantic via pgvector), and delivery across 7 messaging channels. It targets teams running AI agents at scale who want to avoid building the plumbing themselves. The desktop 'Lite' edition (SQLite, Wails/React, max 5 agents) is a lower-friction entry point for individuals.
The single-binary deployment story is real: ~25 MB, no Node.js runtime, sub-second startup, runs on a $5 VPS. For teams tired of managing Python venvs and sidecar processes, that matters. The 3-tier memory design (working → episodic summaries → semantic knowledge graph with pgvector) is architecturally coherent rather than just a vector store bolted onto a chat loop — progressive L0/L1/L2 loading avoids stuffing the full graph into every context window. Go's goroutine model is a genuine fit for the concurrency requirements here: parallel agent delegation, async tool execution, and dedup/retry worker pools are things Go handles without ceremony. The built-in LLM call tracing — spans with prompt cache metrics, optional OTLP export — is the kind of observability that most agent frameworks treat as an afterthought.
The license is CC BY-NC 4.0, which means you cannot use this in any commercial product or service without a separate agreement. That rules out the majority of real-world deployments before you write a line of code. The 'self-evolution' feature — agents auto-adapting their own communication style and CAPABILITIES.md via a guardrailed pipeline — is an operational liability in production; when something goes wrong, 'the agent changed itself' is a very hard incident to debug or roll back. The scope is enormous (30+ tools, 7 channels, 20+ providers, knowledge graph, RBAC, sandbox, cron, webhooks, desktop app), and that breadth almost always means some of these integrations are thin wrappers that will fail in non-obvious ways under real load. The fork count (937 forks, 3322 stars) is a 28% fork-to-star ratio that's unusually high and worth scrutinizing before building anything on top of this.