anda-hippocampus

OpenClaw plugin for Anda Hippocampus that gives agents persistent long-term memory.

It does two things:

• automatically sends completed conversations to Anda Hippocampus for memory formation
• registers a recall_memory tool so the agent can retrieve relevant memory with natural-language queries

By default the plugin uses the hosted Anda Hippocampus service, but baseUrl can point to a self-hosted or local deployment when memory data must stay inside your own network.

What This Plugin Provides

| Capability | Type | What it does | | --------------- | -------------- | ------------------------------------------------------------------------------------------------ | | agent_end | Automatic hook | Converts the finished conversation into Hippocampus messages and sends them to the formation API | | recall_memory | Agent tool | Queries long-term memory using natural language and returns the synthesized result |

This plugin does not try to replace OpenClaw's full prompt lifecycle. It focuses on one reliable write path and one recall tool.

Prerequisites

• OpenClaw >=2026.3.8
• An Anda Hippocampus space ID and space token
• Optional: a self-hosted or local Anda Hippocampus deployment if you do not want to use the hosted service

Installation

The plugin is auto-discovered by OpenClaw from the package metadata. After installation, you only need to enable it in openclaw.json.

1. Install from SKILL.md via Claw

Please add plugin "anda-hippocampus" from https://brain.anda.ai/SKILL.md to your OpenClaw workspace. Use the following configuration:

{
  "spaceId": "my_space_001",
  "spaceToken": "ST_xxxxx",
  "baseUrl": "https://brain.anda.ai" // hosted default
  // "baseUrl": "http://localhost:8042" // self-hosted/local deployment
  // ... other optional config fields ...
}

2. Install the plugin via CLI

1. Install the plugin package:

openclaw plugins install anda-hippocampus

2. Update anda-hippocampus configuration in openclaw.json:

{
  "plugins": {
    "entries": {
      "anda-hippocampus": {
        "enabled": true,
        "config": {
          "spaceId": "my_space_001",
          "spaceToken": "ST_xxxxx",
          "baseUrl": "https://brain.anda.ai" // or "http://localhost:8042" for self-hosted/local
        }
      }
    }
  }
}

3. Restart OpenClaw Gateway.

openclaw gateway restart

Required fields:

• spaceId: your Hippocampus space ID
• spaceToken: your space token
• baseUrl: optional, defaults to https://brain.anda.ai; set this to your own deployment for self-hosted or local use

If memory data must stay inside your network, set baseUrl to your own Anda Hippocampus deployment.

Configuration

| Option | Type | Required | Default | Description | | -------------------- | -------------- | -------- | ----------------------- | -------------------------------------------------------------------------------------------- | | spaceId | string | Yes | — | Memory space ID used for both formation and recall | | spaceToken | string | Yes | — | Bearer token used to authenticate API requests | | baseUrl | string | No | https://brain.anda.ai | Anda Hippocampus base URL. Can point to the hosted service or a self-hosted/local deployment | | defaultContext | InputContext | No | — | Context merged into every request | | formationTimeoutMs | number | No | 30000 | Timeout for formation requests | | recallTimeoutMs | number | No | 120000 | Timeout for recall requests |

defaultContext

interface InputContext {
  user?: string
  agent?: string
  session?: string
  topic?: string
}

Notes:

• defaultContext is merged with per-call recall context
• during agent_end, the plugin fills agent and session from the OpenClaw runtime when available
• baseUrl has trailing slashes removed automatically

How It Works

Agent turn completes in OpenClaw
        |
        v
`agent_end` hook extracts user and assistant messages
        |
        v
POST /v1/{space_id}/formation
        |
        v
Anda Hippocampus queues background memory formation

Later, when the agent needs memory:
        |
        v
Agent calls `recall_memory`
        |
        v
POST /v1/{space_id}/recall
        |
        v
Plugin returns the synthesized memory result to the agent

Formation Behavior

After each completed agent turn, the plugin converts AgentMessage[] into Hippocampus messages and sends them to:

POST /v1/{space_id}/formation

Formation is fire-and-forget. The plugin does not block the agent while the service processes memory in the background.

Current conversion behavior:

• user text messages are included
• assistant text content is included
• tool results and custom messages are skipped
• message timestamps are preserved when available

Recall Tool

The plugin registers a recall_memory tool that sends queries to:

POST /v1/{space_id}/recall

Recall may take longer than normal tool calls because the Hippocampus service may search a knowledge graph and synthesize a response. The default timeout is 120 seconds.

Tool parameters:

| Parameter | Type | Required | Description | | --------------- | -------- | -------- | ----------------------------- | | query | string | Yes | Natural-language memory query | | context.user | string | No | Current user identifier | | context.agent | string | No | Calling agent identifier | | context.topic | string | No | Topic hint for disambiguation |

Example:

{
  "query": "What preferences has Alice expressed about release communication?",
  "context": {
    "user": "alice",
    "topic": "product launches"
  }
}

Operational Notes

• Formation failures are logged and do not crash the agent flow
• Recall failures are returned as tool errors so the agent can react explicitly
• If no relevant memory is found, the tool returns No relevant memory found.

Troubleshooting

| Problem | Likely cause | What to check | | ----------------------------------- | --------------------------------------------- | ---------------------------------------------------------------------------------- | | Formation requests are not arriving | Invalid spaceId, spaceToken, or baseUrl | Verify credentials and watch for [anda-hippocampus] Formation failed logs | | Recall times out | Recall search is taking too long | Increase recallTimeoutMs | | Empty recall results | Query too vague or memory not formed yet | Try a more specific query with context.user, context.agent, or context.topic | | Data must stay on-prem | Hosted default not acceptable | Point baseUrl to your self-hosted or local deployment |

License

Copyright © LDC Labs

Licensed under Apache-2.0 license.