Lighthouse vs Walrus

Lighthouse combines decentralized file storage with a long-term memory service for AI agents. Lighthouse Memory saves project context, preferences, and decisions for retrieval in later sessions, with remember, recall, and forget operations. Its batched engine supports local hybrid retrieval using embeddings alongside keyword or tag matching. Memories are documented as verifiable, content-addressed blobs, and the hosted backend exposes tools through the Model Context Protocol for Claude Code, Claude Desktop, and other compatible agents. For files, Lighthouse uses IPFS and lists Filecoin and Walrus as storage networks; its web app supports storing, retrieving, and sharing files. Developers can use an SDK and CLI, while documented memory storage options include Lighthouse Walrus, Lighthouse IPFS/Filecoin, AWS S3, R2, MinIO, and a local folder. Plans include a free tier with 5 GB and 10,000 AI memories, Lite at $12 a month for 500 GB and 25,000 memories, and Premium at $49 a month for 2.5 TB and 100,000 memories. Encryption and other features are add-ons. Lighthouse cautions that forgetting a memory may not erase its stored blob, and content identifiers do not establish whether content is true.

Walrus is a verifiable data platform for storing and managing data across apps, providers, and agents. Its decentralized storage protocol is coordinated onchain through Sui. Storage costs $0.023/GB/month, paid in WAL, with SUI used for transaction gas; stored data expires when its paid duration ends. Walrus Memory is intended to give AI agents persistent, portable memory across apps and sessions. It supports semantic remember and recall, automatic fact extraction, and generated answers grounded in stored memories. Listed integrations include native MCP support, plugins for OpenClaw and NemoClaw, and compatibility with major LLMs. Developers get TypeScript and Python SDKs, a relayer, smart contract, indexer, and dashboard. Memories are encrypted before storage and decrypted only for an authorized signed request; Sui contracts enforce ownership and programmable access, including scoped delegation. The separate Walrus Console is in a closed, invite-only Mainnet beta, with each space limited to 5 GB and five folders and uploads capped at 100 MiB. Walrus Memory has a launch-period free plan, with usage limits that may change.