Replace GoClaw memory system with OpenViking context database

Proposal

Replace the existing GoClaw memory implementation with OpenViking — an open-source context database designed specifically for AI Agents by ByteDance/Volcengine.

Why OpenViking?

1. Unified Context Management

OpenViking replaces fragmented memory storage (memories in code, resources in vector DBs, skills elsewhere) with a filesystem-based paradigm where developers can organize an agent's brain like managing local files. This eliminates the scattered context problem GoClaw currently faces.

2. Reduced Token Consumption & Cost

OpenViking implements an L0/L1/L2 three-tier context loading structure that loads context on demand. This significantly reduces token usage compared to traditional flat memory approaches — directly cutting API costs.

3. Superior Retrieval Accuracy

Directory recursive retrieval combines directory positioning with semantic search, achieving more precise context acquisition than flat-storage RAG systems. Traditional RAG lacks a global view and creates implicit "black box" retrieval chains — OpenViking solves this.

4. Transparent Debugging

Visualized retrieval trajectories let developers observe retrieval paths, making debugging transparent and enabling optimization of retrieval logic. No more guessing why the agent pulled irrelevant context.

5. Self-Evolving Memory

Automatic session management compresses conversations and extracts long-term memory automatically, allowing agents to improve with continued use without manual intervention.

6. Hierarchical Context Delivery

Structured directory organization enables hierarchical context delivery — agents get the right context at the right granularity level, rather than dumping everything into a flat vector store.

Benefits Summary

References

• OpenViking repo: https://github.com/volcengine/OpenViking/
• License: Open source (Apache 2.0)

Next Steps

• Evaluate OpenViking integration effort with current GoClaw architecture
• Benchmark token consumption: current memory vs OpenViking tiered loading
• Prototype migration path for existing memory data

No files, tests, or entry points are named. Start by locating the current GoClaw memory implementation and comparing its architecture with OpenViking. Done requires an integration-effort evaluation, token-consumption benchmarks, and a documented migration prototype for existing memory data.