ClaudeClaw gives Claude Code persistent memory across sessions
Open-source plugin adds three-tier memory, always-on channel listening, and webhook triggers to Claude Code agents.
ClaudeClaw is an open-source orchestration layer that upgrades Claude Code from a one-shot tool into a persistent agent loop. It listens across Slack, Telegram, and WhatsApp, batching context between triggers and routing messages to isolated Claude agents per group.
Memory is structured in three tiers: daily append-only logs, topic-specific files, and long-term CLAUDE.md files with a memory_search tool for semantic recall. Transcripts are archived before context compaction — nothing gets lost between sessions. Agents run in OS-level sandboxes (Anthropic's sandbox runtime) with kernel-enforced filesystem and network isolation, cold-starting in under 10ms. Webhook endpoints with HMAC-SHA256 auth let external systems — CI/CD pipelines, GitHub events, monitoring alerts — trigger agents directly.
Setup is interactive: run /setup inside Claude Code and it configures dependencies, runtimes, and channels itself. The entire codebase fits in a single context window (~35K tokens), so you can ask Claude to modify it on the fly.
Why it matters: persistent memory has been one of the biggest gaps in agentic Claude workflows — ClaudeClaw patches it without waiting for Anthropic to ship it natively.
Sources
- ClaudeClaw on GitHub github.com
Written by an AI pipeline from the sources above. How it works.
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.