Signed installers on macOS and Windows, a daily encrypted backup on your own computer, and a written list of what every 1.x release keeps.
Atlanta, GA · November 17, 2026 · For immediate release
NexSpark LLC today released ModelBrain 1.0, a memory layer for AI assistants that runs on the user's own computer. Assistants that support the Model Context Protocol (MCP), including Claude Desktop, Claude Code, Cursor and ChatGPT, share one memory, so something saved in one is available in the others. The local plan is free and needs no account.
For a memory tool, 1.0 is a promise about what will not change. ModelBrain 1.0 publishes that promise as a list on What 1.x promises: the store format and the settings assistants rely on stay the same through every 1.x release, and upgrading re-indexes nothing. Each release is checked against that list before it ships. Stores saved by earlier releases, going back to 0.5.1, are kept as test fixtures and must open unchanged, and a store written by 1.0 still opens in 0.6.8 for anyone who needs to go back.
The same check covers results. ModelBrain's published benchmark figures came from 0.6.8, and each 1.0 build was compared with them question by question: on all 1,942 test questions from LoCoMo, BEAM and LongMemEval, 1.0 returns the same five results in the same order. On those test sets ModelBrain is ahead of Mem0, Honcho and total-agent-memory in 31 of 32 retrieval comparisons and tied in 1, and it answers 320 of 370 LongMemEval questions correctly, against 312 for Mem0, 302 for Honcho and 299 for total-agent-memory. The method and caveats are on the benchmarks page.
1.0 also carries the install and data protection work that arrived in its release candidates, 0.9.0 and 0.9.2. The Windows installers and the 14 ModelBrain programs they install are signed by NexSpark LLC (the small uninstaller that Inno Setup writes is not), and uninstalling on Windows stops the background service and removes ModelBrain from Claude Desktop, Cursor and Claude Code before it removes the files. ModelBrain keeps a daily encrypted snapshot of the memory store on the user's own computer, seven at a time, and a restore that uses the wrong key says so and says where the daily backups keep their key. When a disk fills up, saves stop with a clear message while recall keeps working, and saving resumes on its own once there is room.
"Your memory is the one part of an AI setup you can't download again. 1.0 means we've written down what we won't break, and every release has to pass that list before it ships."
Matthew Firth, founder of ModelBrain
Before release, ModelBrain 1.0 was installed over the previous release on macOS, on Windows 11 on Arm in a virtual machine, and on Linux, run through a full disk, and restored from backups, including one made a month earlier. On a copy of a real store, recall answered in under 900 ms at the 95th percentile over 150 searches, the Claude Code hook in about 10 ms, and saves in under 20 ms.
ModelBrain 1.0 is available now on the download page. It is also listed in the official MCP Registry as net.modelbrain/mcp, and the launcher is on npm as modelbrain-mcp. An optional open source plugin, listed in Obsidian's plugin directory, searches an Obsidian vault and saved memories in one panel and never changes a note. It has been tested on macOS so far. Plugin guide.
NexSpark LLC is the company behind ModelBrain, a second brain for AI assistants that runs on your own computer.
Matthew Firth, hello@modelbrain.net
Archived releases, newest first, each superseded by the release above it.
Sync your AI memory across computers with managed backup, or share it across a team. The free local plan stays free.
Atlanta, GA · October 11, 2026 · For immediate release
NexSpark LLC today announced paid device sync with managed backup and team plans for ModelBrain, a memory layer for AI assistants that runs on the user's own computer. Assistants that support the Model Context Protocol (MCP), including Claude Desktop, Claude Code, Cursor and ChatGPT, share one memory, so something saved in one is available in the others.
Until now ModelBrain kept memory on a single computer. Device sync keeps the same memory on each of a person's computers and backs it up. Team plans let a group share a memory across its members' computers. Device sync costs $6 per device per month billed yearly, or $8 monthly, plus $0.05 per GB of storage per month. Team plans are $6 per node per month. Payments are handled by Link, which acts as merchant of record.
The free plan is unchanged. For local use there is no account and no cost, and saving and recalling make no server calls apart from the one-time download of the optional reasoning model. Memories are stored as ordinary files on the user's machine and are never used as training data. ModelBrain can also read the relevant notes with a language model that runs on the user's own hardware, answer the question and cite which notes it used. The model downloads once, at roughly 1.5 to 2.5 GB, then runs locally with no API key.
"People switch between Claude Desktop, Claude Code, Cursor and ChatGPT during a single day, and each one starts from zero. Sync extends the same idea to your second laptop and your team: one shared memory, wherever you work."
Matthew Firth, founder of ModelBrain
ModelBrain is available for macOS (Apple Silicon) and Windows (x64 and ARM64). A command-line package for Linux is also available. ChatGPT connects through a relay and needs a free ModelBrain account. The Windows installers are not yet code-signed, so Windows SmartScreen shows a warning on first run.
ModelBrain publishes how it tests its own recall. In a 269-question evaluation, each question run three times, it gave the right answer 86% of the time by a lenient automatic match and 70% by a strict exact-match standard. It scored 100% on prompt-injection resistance and 79% on its weakest category, questions that need two separate notes. The method and limits are on the results page.
ModelBrain is available now on the download page. It is also listed in the official MCP Registry as net.modelbrain/mcp, and the launcher is on npm as modelbrain-mcp.
NexSpark LLC is the company behind ModelBrain, a second brain for AI assistants that runs on your own computer.
Matthew Firth, hello@modelbrain.net
A model sized to your computer reads your notes and cites its sources. ModelBrain now runs on macOS, Windows and Linux and works with Claude Desktop, Claude Code, Cursor and ChatGPT over the Model Context Protocol.
Atlanta, GA · October 6, 2026 · For immediate release
NexSpark LLC today announced the release of ModelBrain, a free memory layer for AI assistants that runs on the user's own computer. Assistants that support the Model Context Protocol (MCP) can remember, recall, expand and forget information through the same local store, so something saved in one assistant is available in the others.
This release adds local reasoning. Instead of only returning matching notes, ModelBrain can read the relevant notes with a language model that runs on the user's own hardware, answer the question, and cite which notes the answer came from. A model sized to the computer downloads once, at roughly 1.5 to 2.5 GB, and then runs locally with no API key and no per-question cost.
ModelBrain stores memories as ordinary files on the user's machine. For local use there is no account to create and nothing is sent to a server. Memories are never used as training data.
"People switch between Claude Desktop, Claude Code, Cursor and ChatGPT during a single day, and each one starts from zero. ModelBrain gives them one shared memory over MCP, so a decision you record in Cursor is there when you open Claude Code an hour later."
Matthew Firth, founder of ModelBrain
ModelBrain registers itself with Claude Desktop automatically, offers one-button setup for Claude Code and Cursor, and connects to ChatGPT through an OAuth relay. The macOS build for Apple Silicon is notarized and signed. The Windows installers for x64 and ARM64 are not yet code-signed, so Windows SmartScreen shows a warning on first run. The Linux package is a command-line install for x86_64 and ARM64 on glibc 2.28 or newer (Debian 10+, Ubuntu 20.04+, RHEL 8+) that registers with Claude Code and Cursor. The x86_64 package was tested on Ubuntu 20.04.
ModelBrain publishes how it tests its own recall. In a 269-question adversarial evaluation, run three times (807 scored answers) and averaged against the production prompt path, with pass and fail criteria written in advance, it scored an 86% right-answer rate and 100% on prompt-injection resistance. The same evaluation reports a 70% strict accuracy figure and a 79% score for its weakest category. The method and results are on the fact sheet and in What "accurate" should mean for an AI's memory.
ModelBrain is available now on the download page. It is also listed in the official MCP Registry as net.modelbrain/mcp, and the launcher is on npm as modelbrain-mcp.