Summary
An agent can recall the exact sentence it was told, not something like it.
Memory is addressed rather than searched, so recall returns the stored text verbatim with its source. The shared pool is the other half: several agents write into the same memory, tagged by speaker, and read each other's.
What it is
A local memory store for AI agents. Content is stored under an identifier and retrieved by direct addressing rather than vector similarity, so what comes back is the text that was indexed, with its attribution. It runs on your machine, and storage is local and encrypted.
What you get
- Content stored for later recall, with a source and optionally the agent that said it
- Memory queried for relevant context, filterable by which agent wrote it
- Files or whole directories imported in one call
- Git commit history imported as memory
- Session summaries stored, and conversation turns saved so a new session picks up where the last stopped
- Project state tracked, and code patterns kept as reusable snippets
- A health check on the server
Requirements
A download and install from solonai.com — install.sh on macOS or Linux — then a restart of the AI tool you use it from. No account and no key; nothing is sent off the machine.
Setup effort
One command — docker pull ghcr.io/solonai-com/grantai-memory:1.8.6