Jasper Recall

What This Skill Does

Jasper Recall is a robust local Retrieval-Augmented Generation (RAG) system designed specifically for the OpenClaw AI ecosystem. By leveraging ChromaDB as a vector database and sentence-transformers for local embedding generation, this skill enables your agents to maintain long-term, searchable memory. It bridges the gap between ephemeral chat sessions and persistent knowledge by digesting conversation logs, indexing them, and providing a semantic interface to retrieve past decisions, technical patterns, and user preferences. Version 0.3.1 introduces advanced features like multi-agent mesh, which allows groups of agents to share memory pools, and specific collection isolation to ensure that private, shared, and learning-based data remains categorized and accessible only where appropriate.

Installation

Installation is streamlined through the OpenClaw CLI and NPM. To set up the skill, run the following command in your terminal: npx jasper-recall setup. This process creates a dedicated Python virtual environment at ~/.openclaw/rag-env, ensuring that all heavy-duty dependencies—such as the vector search engine and embedding models—are isolated from your primary system libraries. It also initializes the ChromaDB storage at ~/.openclaw/chroma-db. Because the system runs locally and requires no external API keys, all data remains on your machine, adhering to strict privacy standards.

Use Cases

Jasper Recall excels in scenarios where context preservation is critical. You can use it to build a searchable knowledge base of project decisions, ensuring that "how" and "why" questions regarding your codebase are answered based on previous developer interactions. It is ideal for session continuity, allowing an agent to "remember" a task from a restart days ago. Additionally, it supports multi-agent workflows where one 'researcher' agent can index information that an 'executor' agent later recalls for implementation.

Example Prompts

1. "recall 'what was the final decision regarding the authentication middleware architecture from our last meeting?'"
2. "index-digests --since 2023-10-01"
3. "recall 'common patterns found in the hopeIDS documentation' --limit 5"

Tips & Limitations

The system is highly powerful but does require significant disk space for the vector database and embedding models (~200MB). When managing multi-agent setups, ensure you are using the correct collection names to avoid data leakage between private and public workspaces. For Docker-based deployments, utilize the serve command to expose the recall API, as the CLI tool may have restricted access to the host's file system within isolated containers. Regularly run digest-sessions to ensure your most recent conversations are moved from raw logs into the searchable index.