A memory layer where agents curate knowledge. Episodes in, curated concepts out — with first-class fact handling and conflict detection.
Your agent decides what to remember and how to organize it. Remind is the substrate, not the intelligence.
Build any workflow on top of Remind — write your own skills for any use case.
Skills use project-local databases (.remind/ in your repo). MCP uses a centralized server. Pick what fits.
Queries activate matching concepts, which activate related concepts through the graph. Like how human memory works.
No API keys needed. Local embeddings via fastembed. Optional remote providers for higher quality.
Dashboard, concept graph, entity explorer, and conflict inbox — all built in.
Entity graph showing concepts, tools, and their relationships.
Your agent curates memory. Remind handles storage, clustering, and retrieval.
Store experiences as episodes. Fast — uses local embeddings by default.
remind remember "User prefers Rust for systems work"
remind remember "Cache TTL is 300s" -t fact -e tool:redisYour agent reviews pending state and organizes knowledge via batch tools.
remind snapshot pending,conflicts
remind apply << 'EOF'
concept from=ep:11,ep:12 title="Tech stack" "Rust + Redis"
processed ids=ep:11,ep:12
EOFSpreading activation retrieval — not just keyword matching. Time-travel for facts.
remind recall "What tech stack decisions have we made?"
remind recall --as-of 2024-06-01 "cache config"Returns curated concepts with provenance, not raw transcripts.
Persistent context across coding sessions. Your agent remembers preferences, decisions, and project architecture.
Ongoing debates across sessions. Remind tracks arguments, positions, and open threads.
Feed in papers, find commonalities and contradictions. The agent surfaces themes across sources.