Production Memory Infrastructure for AI Agents

HindClaw is production memory infrastructure for AI agents. It centers on server-side access control, Terraform-managed banks, and multi-agent memory with per-user permissions, and it is built on Hindsight by Vectorize. The project describes itself as three independent packages that can be used separately or together.

Its server-side extension provides JWT authentication, permission enforcement, tag injection, and a management REST API, and it can be installed on any Hindsight server. The Terraform provider manages users, groups, banks, permissions, directives, mental models, and entity labels as code, with the documentation noting that terraform apply makes the stack live. The OpenClaw gateway plugin is described as a thin adapter that signs JWTs and auto-starts the embed daemon with extensions loaded.

The product’s workflow routes a message from channels such as Telegram or Slack to the right agent, signs a JWT with sender, agent, channel, and topic context, resolves the sender to a user, checks group memberships, and applies a four-layer permission cascade. Memory operations then run server-side: recall returns filtered results with tag groups, while retain stores data with injected tags and strategy. Additional features listed on the page include per-agent memory banks, multi-bank recall with per-bank permission checks, named retain strategies, and entity labels with multilingual aliases, tag generation, and graph-traversable entities.

The page also says HindClaw is built on Hindsight by Vectorize and contrasts that system’s memory engine with HindClaw’s access-control and infrastructure layer. Delivery is through the Hindsight server extension, the Terraform provider, and the npm-published OpenClaw gateway plugin. A link to PyPI appears for the extension, and the site also references Hindsight Cloud as a managed option. No pricing or license terms are stated in the provided text.