What is hindclaw?
hindclaw is a production-grade Hindsight memory system for OpenClaw. It consists of two components: a thin plugin that generates JWTs and sends standard Hindsight API calls, and a server extension (hindclaw-extension) that enforces access control, resolves permissions, and enriches requests with tags and strategies. Together they give your AI agent fleet long-term memory with per-agent configuration, multi-bank recall, named retain strategies, and infrastructure-as-code management.
Two Dimensions
Every message resolves along two orthogonal axes:
WHO -- permissions enforced server-side by the hindclaw-extension. Users, groups, and bank-level overrides are managed via the /ext/hindclaw/* REST API (not config files). Resolution follows 4 layers (global group defaults -> bank _default baseline -> bank group overlay -> bank user override) with 11 configurable fields at every layer: LLM model, token budget, extraction depth, tag visibility, retention frequency.
HOW -- strategy resolved per scope via a 5-level cascade (agent -> channel -> topic -> group -> user). Each scope routes to a named strategy with its own extraction mission, mode, and entity labels. Strategy resolution and tag enrichment happen server-side.
Permission resolution, tag injection, and strategy selection all happen server-side in the hindclaw-extension. The plugin is a thin adapter that generates a JWT and sends standard Hindsight API calls. Every combination of (user x bank x topic) can produce different behavior.
Core Features
Per-agent bank configs -- each agent gets its own retain mission, entity labels, dispositions, and directives. Configured via JSON5 files, synced to Hindsight via hindclaw apply.
Multi-bank recall -- agents read from multiple banks in parallel. A strategic advisor recalls from finance, marketing, and ops banks simultaneously.
Named retain strategies -- map conversation topics to extraction profiles. Strategic conversations get deep analysis, daily chats get lightweight extraction. Strategies can also be assigned per user group.
Access control -- server-side via hindclaw-extension. Users belong to groups. Groups define defaults. Banks override per-group or per-user. Anonymous users blocked by default. Dual authentication: JWT for plugins acting on behalf of users, API keys for direct access (CLI, dashboards, personal tools). Users and groups are managed via the /ext/hindclaw/* REST API.
Infrastructure as Code -- hindclaw plan/apply/import. Declare bank configs in JSON5 files, diff against server state, apply changes. Like Terraform for memory banks.
Session start context -- mental models loaded before the first message. No cold start.
Reflect-on-recall -- use Hindsight's reflect API instead of raw recall for richer, reasoned responses.
Multi-server -- per-agent infrastructure routing. One gateway, multiple Hindsight servers (home, office, local daemon).
Built on Hindsight
Hindsight is a biomimetic memory system for AI agents with semantic, BM25, graph, and temporal retrieval. hindclaw is a client that maps OpenClaw concepts (agents, channels, topics, users) onto Hindsight capabilities (banks, strategies, tags, tag_groups).
Next Steps
- Installation -- set up the plugin and (optionally) the server extension
- Bank Configuration -- configure your first agent's memory
- Access Control -- set up multi-user permissions via hindclaw-extension
- Configuration Reference -- plugin and JWT configuration