Save what your agent needs to know next time.
Flanner Memory stores the decisions, constraints and facts around your project — why you chose a database, which approach was rejected, how the staging environment works — as Markdown files that any later session can search.
How it works
Stored as Markdown files
Each memory is a file under .flanner/memory/ in your repository. You can read it, diff it, and commit it. The database is only an index over those files, so if you lose it, flanner mem rebuild restores it from the files.
Searchable from any session
flanner mem recall searches the full text and shows why each result matched. An agent starting a new session next week can find a constraint you saved today.
Corrected, not overwritten
Superseding a memory keeps the old one readable and links the new one to it. You can see what changed and when.
Explicit by default
Nothing is saved unless you ask. In suggest mode, your agent proposes memories and you approve them. auto-safe mode exists, ships off, and can only be enabled in a project's own policy file.
Save once, find it later
The whole product in eight lines. The last line is what recall adds over grep: it tells you why the result matched.
Also: flanner mem mode suggest lets your agent propose memories for you to approve, flanner mem tag labels them by topic so related ones are easy to find, and flanner mem rebuild restores the index from the files. All three are in the docs.
Sharing with your team
Explicit, signed, and device to device.
With Flanner Mesh, you can share a memory with your workspace by running flanner mem share. Sharing is always explicit: joining a workspace does not share any existing memories.
A shared memory is signed by the machine that wrote it and travels directly to your teammates' devices. It never passes through Flanner's servers. The receiving device verifies the signature and runs the same secret check before storing it.
flanner mem withdraw asks every device to stop recalling a shared memory. It does not delete the file from a device that already received it. How sharing works.
What it will not do
It will not store secrets
Before anything is written, the text is checked against fifteen vendor key patterns, a generic credential check, and an entropy check. This runs on what you type, on what an agent proposes, and on anything a teammate sends you.
It will not act on memories
A memory is context an agent reads, not an instruction it follows. Personal memory — notes about how you work — is never shared with anyone.
Memory is free, in the open-source client
No account, and your memories never leave your machine.
uv tool install flanner