Documentation · MemoryAgent

Install the command-line client

macOS and Linux. The installer fetches the published manifest, downloads the bundle, verifies it against the manifest's sha256, and installs a launcher. It compiles nothing and needs no root.

install.sh

curl -fsSL https://pfmem-api.packagefactory.dk/install.sh | sh

Then put the launcher on your PATH and store a key:

pfmem

export PATH="$HOME/.pfmem/bin:$PATH"
pfmem init

Where to go

Quickstart
Sign up, mint a key, create a workspace, install the client, store a conversation, read it back. Seven steps, and only one of them cannot be repeated.
Read on
CLI
Install, the stdout contract a harness parses against, every command and flag, the exit codes, the log file, updates, and the skill an agent reads before its first call.
Read on
HTTP API
The base URL, the two ways to authenticate, the response envelope, the status codes, the key tiers, and what every endpoint does.
Read on

Three things that shape every page here

Your key decides what you can see.

A key resolves server-side to an account, a workspace, a role and a set of verbs. Nothing you send names a workspace, a filter or an embedding version — there is no field for it, in the API or in the client. That absence is what keeps one customer's memories out of another customer's results.

Reads are POST too.

Nearly every endpoint is a POST, reads included, because a read takes a body with filters in it and a key belongs in a header rather than a URL. The two exceptions are the task endpoints, which are GETs. This is deliberate.

Timestamps are unix milliseconds.

Thirteen digits, at or above 1000000000000. A seconds-precision value is rejected outright rather than quietly read as a date in 1970.