fortemi-react — March 2026

fortemi-react is the Fortemi memory service, built to run inside your web browser. Fortemi keeps notes for an AI agent and lets the agent search them by meaning. fortemi-react does that same job with no separate server — it all runs in the browser, on your computer. (It is the server in the browser. HotM, a different project, is a client that talks to the server.)

TL;DR

This is where fortemi-react started. The project began on March 20, 2026, and in its first days it went from an empty repository to a working memory system that runs entirely in the browser. By the end of the month you could keep notes, search them by keyword and by meaning, let on-device AI write titles and tags, attach files, and pack a whole set of notes into a single shareable file. Nothing was published to the public npm registry yet — the first public packages came in May — but the whole foundation was built this month. This is an inception report: a look at what got laid down.

By the numbers

What's public	Value
Project started	March 20, 2026
npm packages	none yet — first public release shipped in May
Where it runs	entirely in your web browser, no server
Source	github.com/Fortemi/fortemi-react

Highlights

1. A memory system that runs in the browser — no server. What it is: fortemi-react keeps an AI agent's notes and lets it search them, and it does the whole job inside a normal web browser. There is no separate server to run. How you'd use it: open a page, and your notes live right there on your computer. Why it helps: your notes stay on your machine, private by default, with nothing to install or host.

2. A real database, inside the page. What it is: notes are stored in a full database (Postgres) that was made to run inside the browser (this build is called PGlite). How you'd use it: you don't manage it — it just works when the page loads. Why it helps: you get the power of a real database — reliable storage, proper search — without any setup or a server.

3. Search by keyword and by meaning. What it is: you can find notes two ways — by the exact words in them, and by what they mean. Meaning-based search turns your words into numbers the computer can compare. This is called an "embedding" — a meaning fingerprint. So a search for "car" can also find a note about "automobile." How you'd use it: type a search; results come back ranked, with short snippets. Why it helps: you find the right note even when you don't remember the exact words.

4. On-device AI helpers. What it is: fortemi-react can write a title for a note, suggest tags, and keep a history of edits — using AI models that run on your own device. How you'd use it: add a note, and the helpers fill in the details in the background. Why it helps: less busywork, and the AI runs locally, so your notes don't have to leave your computer.

5. A shareable "knowledge pack" format. What it is: a whole set of notes can be packed into a single file and later unpacked somewhere else. This is the Knowledge Shard format. How you'd use it: export your notes as one file, hand it to someone, and they import it. Why it helps: notes become portable — easy to back up, move, or share.

Features shipped

The foundation. The project started from nothing on March 20. Within the month, it grew into a working browser memory system. Several pieces came together. There is the code package other apps will use, `@fortemi/core`, inside a shared workspace. There is the in-browser database and the tools that read and write notes. There is a small demo app to try it all. And there is a set of tools an AI agent can call to capture and look up knowledge.

Search, done properly. Full-text (word-by-word) search landed alongside meaning-based search, plus a mixed mode that uses both. Results are ranked, and each comes with a short snippet so you can see why it matched.

The on-device AI pipeline. fortemi-react learned to turn note text into meaning fingerprints, called embeddings, using a model that runs in the browser. It also learned to run a small language model on your device for writing titles and tagging notes. A background job queue handles this work, so the screen stays usable while it runs. Heavy AI models are opt-in. The app checks what your device can do, including its graphics chip, and only loads a model when you ask. So nothing large downloads by surprise.

Attachments and file storage. Notes can carry attached files, stored in a dedicated in-browser store, with a tool for managing them.

The knowledge-pack format. Import and export of the Knowledge Shard format arrived, so a whole set of notes can travel as one file — matching the format the Fortemi server uses, so packs stay compatible across the two.

A clear name. Early in the month the project was renamed from fortemi-browser to fortemi-react, settling on the name it carries today.

Fixes

As the first code came together, early rough edges were smoothed. A flash in the search screen was fixed. Tests were aligned with the server-compatible job queue. And test runs were tuned so the in-browser database wouldn't overload the machine running them.

Performance & reliability

From day one, the AI work was pushed onto a background job queue so it never froze the screen. Loading of heavy models was made opt-in and gated on what the device can handle — a reliability choice as much as a speed one, since it avoids large surprise downloads.

Breaking changes & migrations

None this month. This was the first month of the project — there was nothing earlier to break.

Releases

None this month. Nothing was published to the public npm registry in March. The work happened in the open on GitHub, and the first public packages shipped in May.

Dependencies & security

No security alerts came up this month. Heavy AI models are opt-in and only load when you ask — a deliberate choice that keeps the default experience small and avoids pulling in large downloads you didn't request.

Docs & developer experience

Documentation started early. This month brought the project's first design and planning documents and a set of developer guides covering how the pieces fit together and how to use search. The goal from the start was that a developer could read the guides and get going without digging through the code.

Tests & CI

Testing began with the code. The job-queue tests were aligned with the server's behavior, and test runs were tuned so the in-browser database stays within the machine's limits. The automatic checks that run on every change (CI) were set up and hardened toward the end of the month.

Cross-project impact

• fortemi-react is the browser build of the Fortemi server — so from its very first month, its knowledge-pack format was kept compatible with the Fortemi server, so notes can move between the two.
• The design leans on the wider Fortemi family: the same note model, the same pack format, one shared idea running in a new place (the browser).

Known issues & open threads

• This was an inception month. The foundation was built fast, and the months that follow are about hardening it, publishing it, and filling in the pieces.
• Publishing to npm was still ahead. The code was public on GitHub, but packaged public releases were set up later.

What's next

Turn the foundation into something people can install: package the code, publish it to npm, and put a proper release process behind it. Keep the browser build in step with the Fortemi server. Those steps start in the following months.

Appendix

• Published packages: none yet in March — the first public releases came in May.
• Project start: March 20, 2026.
• Source / docs: github.com/Fortemi/fortemi-react · window: March 20–31, 2026.