ShineMnemos gives AI agents a long-term memory they can trust — exact search, verified writes, honest provenance, and durable storage, all behind one MCP server.
This is a sandboxed demo of the memory pattern — type a fact, watch it pass the gates and land in memory, then query it back.
# stored memories (empty — write a fact to begin.)
A standard MCP server your agent points at. Memory tools are exposed as search / add / verify — no SDK rewrite.
Exact hash lookup in ~1 µs and a hybrid vector layer coming. Deterministic, no false positives.
Truth gate + write guard + provenance. Low-quality or poisoned facts never enter memory.
Context Engine compacts history into a reusable prefix, with a stable layout that preserves provider cache.
Op journal confirms every write. If an operation isn't acknowledged, you hear about it.
Like the best dev tools, our open source is the free channel that earns trust. The business is the managed cloud and enterprise tier — not your data.
Follow the MCP guide, then join the early-access waitlist.