Claw Platforms

Comparative review of NanoClaw alternatives in the OpenClaw ecosystem — personal AI agent runtimes that connect to WhatsApp, Telegram, Slack, and other messaging channels.

Evaluated 2026-04-13.

What we need

• WhatsApp as a primary channel
• Self-hosted on generic x86_64 Linux
• Claude or Ollama as LLM backend
• Minimal operational overhead

Platforms

NanoClaw (current)

qwibitai/nanoclaw · Node.js · Channels: WhatsApp, Telegram, Slack, Discord, Gmail

Lightweight alternative to OpenClaw that runs agents in containers. Claude Agent SDK is the execution model — agents are Claude Code instances with skills injected as prompt text. Memory is per-group filesystem (CLAUDE.md). No observability; docs say "ask Claude Code about issues." Skill system requires forking the repo and merging upstream — no shared registry.

Verdict: Works. Small, auditable codebase. But no OTEL, no multi-agent delegation, single orchestrator, Claude-only (with API-compat workaround). The messiness is real.

SwarmClaw

swarmclawai/swarmclaw · Node.js · Channels: WhatsApp, Telegram, Slack, Discord, Teams, Matrix

Multi-agent orchestration with native delegation, durable execution, and task boards. 23+ LLM providers (Anthropic, OpenAI, Ollama, OpenRouter…). Full OTLP export for chat turns, model streams, and tool calls — the only platform here with first-class observability. Skill system supports OpenClaw-compatible SKILL.md import and "conversation-to-skill" drafting with human review gates.

Memory is a hybrid: durable documents, project-scoped context, reflection memory, graph traversal, and communication preferences.

Verdict: The most capable option. OTEL + WhatsApp + 23 LLMs + multi-agent delegation is a strong combination. High operational complexity is the cost.

Moltis

moltis-org/moltis · Rust · Channels: WhatsApp, Telegram, Discord, Teams, Web UI

Single 44MB binary (zero unsafe code, 46 auditable crates, 3,100+ tests). Full OTEL tracing + Prometheus metrics. SQLite + FTS + optional vector embeddings for memory. MCP server support. Sandboxed tool execution. Encryption at rest (XChaCha20-Poly1305 + Argon2id). Multi-arch Docker (amd64/arm64).

LLM support is the weak point — no Anthropic Claude or GPT-4 direct (works through OpenAI-compatible proxy). Single maintainer.

Verdict: Best architecture story. Rust memory safety + OTEL + single binary + WhatsApp. The LLM limitation hurts but may be patchable via ANTHROPIC_BASE_URL-style proxy.

nanobot

HKUDS/nanobot · Python · Channels: WhatsApp, Telegram, Discord, WeChat, Slack, Feishu, DingTalk, QQ, WeCom, Matrix, Email

Widest channel coverage by far — 11+ via plugin system including email and Chinese platforms. 10+ LLM providers including Ollama. ClawHub skill marketplace. MCP native. Dream two-stage memory (v0.1.5+). Hardened against session poisoning.

No native OTEL — LangSmith only. Documentation sparse. Still in beta.

Verdict: Best channel coverage. Good fit if you need breadth. Not ready for production-critical workflows — observability is the gap.

AlexClaw

thatsme/AlexClaw · Elixir/OTP · Channels: Telegram, Discord

OTP supervision trees, circuit breakers, four-tier LLM cost routing (light → heavy → local), PostgreSQL + pgvector for hybrid search memory, dynamic skill loading without restarts. Fully offline capable. Privacy-first — no data leaves your infrastructure.

No WhatsApp. Single-user only.

Verdict: Best architecture for fault tolerance and privacy. The Telegram/Discord-only constraint rules it out as a NanoClaw replacement, but worth watching if WhatsApp support lands.

PicoClaw

sipeed/picoclaw · Go · Channels: Telegram, Discord, WeChat, WeCom, Matrix, IRC

Single binary, <10MB RAM, boots in <1s on 0.6GHz hardware. Runs on RISC-V, ARM, MIPS, x86, Android. 30+ LLM providers. 26K stars. Sub-agent architecture with EventBus and hooks.

Pre-release (v0.x), 95% AI-generated code, "do not deploy before v1.0." No WhatsApp. No OTEL.

Verdict: Impressive portability story but not production-ready and lacks WhatsApp.

Summary

If switching: SwarmClaw for capability, Moltis for architecture.