Recall Get Recall £19

Tell it once. Your local AI remembers the rest.

Stop re-explaining your setup every session. Drop-in memory for Ollama, Claude Desktop, or Cursor (via MCP) — fully local, set up in 20 minutes.

Get Recall £19 See it remember
Works with Ollama Claude Desktop Cursor (MCP)
The point

It carries what you told it into the next session.

recall — session
I run Proxmox at home and write everything in Python.
Saved.
# — session closed, reopened days later —
write me a quick backup script
Here's a Python script for your Proxmox box…
# you never re-told it — it remembered

Representative session. The kit recalls only what you've saved — nothing leaves your machine.

How it works

Four moving parts. Nothing to learn.

01

Save a note

You (or your agent) save a fact — your stack, a preference, a decision.

02

Embed it locally

A local model turns it into a vector. No cloud, no API key.

03

Retrieve by meaning

Later, the most relevant memories are found by similarity — not keywords.

04

Feed your AI

Those memories go to your agent, so it answers like it knows you.

What you get

A kit, not a framework.

  • memory modulesave() / search()The core store. Read it in one sitting.
  • MCP servermemory_save / memory_searchPlugs into Claude Desktop, Cursor, or your own agent.
  • example agentA runnable CLITell it something, restart, watch it remember.
  • README~20-minute setupCopy-paste, start to finish.
Running in ~20 min

Free OSS memory libraries are frameworks you study. Recall is a small, legible kit you run today — that's the £19.

Under the hood

Exactly what you're running.

Language
Python 3.11+
Runtime
Ollama (local)
Models
llama3.2 · nomic-embed-text (swappable)
Storage
SQLite + sqlite-vec
Integration
Model Context Protocol (MCP)
Footprint
~8 GB RAM · ~5 GB disk
Platforms
Windows · macOS · Linux
Licence
Single-user (one-time)
Upgrade path
SQLite → Postgres + pgvector
What it doesn't do

The hard parts are left out, on purpose.

Two problems are deliberately excluded — so the core stays small enough to actually read and trust.

Relevance filtering

It doesn't decide "what's worth remembering" for you. You save what matters.

Contradiction handling

It won't reconcile memories that disagree over time. It stores and recalls — faithfully.

These are genuinely hard problems, and most tools that claim them do them badly. Recall would rather be honest about its edges than pretend.

FAQ

Straight answers.

Does my data leave my machine?

No. Embeddings and storage are fully local — no network calls, no API keys.

Which models does it use?

llama3.2 for chat and nomic-embed-text for embeddings, via Ollama. Both are swappable.

Can I use it in production?

Yes — swap the storage for Postgres + pgvector. The rest of the kit doesn't change.

What's not included?

Relevance filtering and contradiction handling — deliberately, to keep the core legible.

What do I need first?

Python 3.11+ and Ollama installed locally. That's it.

Get Recall

Give your local AI a memory tonight.

Get Recall £19

one-time purchase · instant download · runs fully local