HippoDid

HippoDid is a persistent memory system for AI agents. It is described as giving each agent a persistent character so memories survive across sessions, restarts, and model upgrades, and as a way to keep memory files from being compacted away when context windows fill up. The service is aimed at developers building agents that need long-term memory across sessions and across different models.

A character is a named memory store, intended for an agent, project, or use case. HippoDid stores memories on the server and makes them searchable by clients that speak REST or MCP. It offers semantic search, time decay weighting, and category filtering, and it supports two memory modes: Verbatim for raw conversation history and Extracted for structured knowledge such as entities, relationships, deduplicated facts, and automatic decay. The site also says HippoDid handles the plumbing so users can write the logic, and that it does not require vector database setup or embeddings infrastructure. It also supports a ready-to-use context assembly function for prompt building.

Delivery options listed on the site include Python, TypeScript, and Java SDKs, along with Spring Boot, MCP, and REST API access. HippoDid syncs memory files server-side and refers to local MEMORY.md files as a live backup. It is shown working with Claude Code, Claude Desktop, Cursor, OpenClaw, LangChain, CrewAI, OpenAI Agents, Vercel AI SDK, Claude MCP, and Spring AI. The pricing page states that it uses BYOK, with AI features requiring the user’s own API key and no AI compute charges from HippoDid itself. A free tier is available with 3 characters, 100 memories per character, AI extraction, semantic search, document import up to 50 KB, and one synced file; paid plans include Starter at $4.99 per month, Developer at $29 per month, Business at $99 per month, and an Enterprise option with custom pricing.