Enables AI clients to perform local-first knowledge management using a browser-based PostgreSQL engine with semantic search and SKOS tagging.
Also known as Fortemi/fortemi-react, @fortemi/core
- Stars
- 5
- Tools
- —
- Downloads, 30 days
- 11,694
- Latest version
- —
- Last commit
- Sep 20, 2026
- License
- AGPL-3.0
What is fortemi-react MCP?
fortemi-react enables AI clients to perform local-first knowledge management using a browser-based PostgreSQL engine with semantic search and SKOS tagging.
Summary generated from the project's README
About fortemi-react MCP server
fortemi-react provides a local-first intelligent database stack designed for browser-based applications. It enables developers to build research tools, notebooks, and knowledge management systems where data remains on the user device by default. The system utilizes PGlite and pgvector to run a real PostgreSQL engine entirely within the client, supporting advanced features like full-text search with tsvector and hybrid semantic search using HNSW and BM25 reciprocal-rank fusion.
This solution is built for React 19 environments and requires no backend or external accounts. Developers can integrate it using @fortemi/core and @fortemi/react packages. While data stays local, users can opt into AI capabilities by connecting to local providers like Ollama, LM Studio, or llama.cpp, or by using WebGPU-based inference via transformers.js. The architecture supports portable Knowledge Shard archives in tar.gz format, ensuring users maintain ownership of their structured data without vendor lock-in.
What you can do with it
- 1Build private research workspacesCreate a notebook application where all user notes and research data stay on their own device.
- 2Implement semantic knowledge retrievalEnable users to find information using hybrid search techniques that combine text and vector embeddings.
- 3Manage complex knowledge structuresOrganize information using SKOS schemes, concepts, and inter-note links within a web interface.
- 4Export portable data archivesAllow users to download their entire knowledge base as a checksummed archive for portability.
Written from the project's README. Every use case is checked against a quote from it.
How to install fortemi-react MCP
Formats verified 10 Nov 2026Add to CursorAdd to GooseInstall fortemi-react MCP in Claude Code
$ claude mcp add fortemi-react -- npx -y @fortemi/coreInstall fortemi-react MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"fortemi-react": {
"command": "npx",
"args": [
"-y",
"@fortemi/core"
]
}
}
}Install fortemi-react MCP in VS Code
.vscode/mcp.json
{
"servers": {
"fortemi-react": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@fortemi/core"
]
}
}
}Install fortemi-react MCP in VS Code CLI
$ code --add-mcp '{"name":"fortemi-react","type":"stdio","command":"npx","args":["-y","@fortemi/core"]}'Install fortemi-react MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"fortemi-react": {
"command": "npx",
"args": [
"-y",
"@fortemi/core"
]
}
}
}Install fortemi-react MCP in Codex
$ codex mcp add fortemi-react -- npx -y @fortemi/coreInstall fortemi-react MCP in Codex config.toml
~/.codex/config.toml
[mcp_servers.fortemi-react] command = "npx" args = ["-y", "@fortemi/core"]
Install fortemi-react MCP in Gemini CLI
$ gemini mcp add -s user fortemi-react npx -- -y @fortemi/coreInstall fortemi-react MCP in Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"fortemi-react": {
"command": "npx",
"args": [
"-y",
"@fortemi/core"
]
}
}
}Automated checks
Ran Nov 15, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablefortemi/fortemi-react is public and reachablegithub.com/fortemi/fortemi-react ↗Pass
- Package names this repository as its source@fortemi/core names fortemi/fortemi-react as sourcenpmjs.com/package/@fortemi/core ↗Pass
- Build provenance (Sigstore attestation)Sigstore attestation for @fortemi/corenpmjs.com/package/@fortemi/core/v/2026.9.9#provenance ↗Pass
- Open source license (OSI)AGPL-3.0spdx.org/licenses/AGPL-3.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-09-20github.com/fortemi/fortemi-react ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNo popular package one edit away from @fortemi/corePass
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubfortemi-react
The browser edition of the Fortémi intelligent-database stack — profile-scoped interchange, client-side, local-first, private by default.
Run the Fortémi schema entirely in the browser: real PostgreSQL (PGlite + pgvector), typed React 19 hooks, and profile-scoped Knowledge Shard archives. Build notebooks, research tools, and AI-powered knowledge apps where the data lives on the user's device, search runs locally, and any cloud or AI provider is opt-in and yours to choose — no backend, no account, no lock-in.
pnpm add @fortemi/core @fortemi/react
pnpm dev
What Fortémi Is · Why · Get Started · Packages · Features · AI Capabilities · Documentation · License
What Fortémi Is
Fortémi is an intelligent database for AI-ready applications: normalized data
contracts plus data-science and processing tooling that turn messy
organizational data into searchable, linkable, provenance-aware structures.
The Fortemi server owns the live APIs and Knowledge Shard authority;
@fortemi/core consumes the declared browser profiles, and HotM consumes the
server and Core contracts as an application. Receipt-bound profiles provide a
common interchange surface without implying one schema across every plane.
fortemi-react is the browser edition of that stack. It runs the normalized schema in PGlite/pgvector with profile-scoped interchange — the browser read/write/search core plus an opt-in, sandbox-limited slice of on-device inference. Knowledge management is one use case built on the substrate, not the whole product. What the browser edition uniquely adds: the data never has to leave the device. Local-first and private-by-default stay the headline differentiators.
Fortémi (product) vs @fortemi/core (embeddable library)
@fortemi/core is an embeddable library — the headless data layer other tools build on. (For example, it is AIWG's default index and discovery backend, reading from a static local cache.) A tool that embeds @fortemi/core is not running Fortémi: it is using the library's schema and query surface inside its own process, and the library ships no data anywhere by default. Fortémi the product is the full stack — the server or this browser edition — with its job pipeline, capability system, and portable archives.
Why fortemi-react
Most in-browser data stacks make you choose: capable but cloud-bound, or private but primitive. fortemi-react refuses the trade-off — it runs a full PostgreSQL engine inside the browser, so data never has to leave the device to be searchable, structured, and AI-ready.
- Private by default. Notes live in the browser on the user's device — no server, no sync service, no cloud backup, no account. The only thing that ever leaves the device is a cloud AI call you explicitly opt into.
- Capable, not toy-grade. Real PostgreSQL with
tsvectorfull-text, pgvector HNSW semantic search, and BM25 reciprocal-rank fusion — production-grade retrieval that stays fast as an archive grows, entirely client-side. - AI on your terms. Embeddings and LLMs are opt-in and bring-your-own: run locally (WebGPU, Ollama, LM Studio, llama.cpp, vLLM, Jan) or route to a provider you control. Nothing is wired to a vendor you can't swap out.
- Yours to keep. Knowledge Shard archives (tar.gz, with checksums and a declared portability profile) export the components supported by that profile on demand, and the AGPL-3.0 license keeps the stack open. No lock-in by design.
Use it when you need local-first note storage, semantic retrieval, agent-readable tool functions, or portable archives inside a web application — whether that's a personal notebook, a research workspace, or a product that must keep user data on the user's device.
Quick Start
Install the published packages:
pnpm add @fortemi/core @fortemi/react
# or
npm install @fortemi/core @fortemi/react
Run the standalone app from this repository:
pnpm install
pnpm dev # http://localhost:5173
Minimal React setup:
import { FortemiProvider, useCreateNote, useSearch } from '@fortemi/react'
export function App() {
return (
<FortemiProvider persistence="opfs" archiveName="default">
<Notebook />
</FortemiProvider>
)
}
function Notebook() {
const { createNote } = useCreateNote()
const { data, search } = useSearch()
return (
<main>
<button onClick={() => void createNote({ title: 'Hello', content: 'First note.' })}>
Add note
</button>
<input onChange={(event) => void search(event.target.value)} />
<ul>
{data?.items.map((item) => <li key={item.id}>{item.title}</li>)}
</ul>
</main>
)
}
Packages
| Package | Published | Purpose |
|---|---|---|
@fortemi/core |
npm | Headless data layer: PGlite repositories, migrations, workers, tool helpers, event bus, capability system |
@fortemi/graph |
npm | Framework-agnostic graph add-on: pure layout, filtering, coloring, degree sizing, bounds/fit, neighborhood, and snapshot helpers, plus a GraphController for graph-source selection. No React. Depends on @fortemi/core; consumed by @fortemi/react and JS-only hosts |
@fortemi/react |
npm | React 19 provider, hooks for notes, search, tags, collections, jobs, capabilities, shards, embedding sets, similarity graphs, and communities, plus 2D/3D graph views (GraphView, SigmaGraphView, ForceGraph3DView) |
@fortemi/standalone |
workspace app | Vite application for local development and static deployment |
All packages are versioned together — the npm badges above always show the current release.
Contract Evidence
Fortemi authority run 6393
passed the same pinned authority-to-consumer contract on Linux x86_64, Linux
arm64, and macOS arm64. The matrix binds React/Core commit
ccf96fad6025025293e40e250c85f088c8999d86 to
@fortemi/core@2026.7.14 (tgz SHA-256
e282f504a842261c3f598a7f2ee0d6a85e03dc213ddf545a18daf5f603a742cc;
tar payload SHA-256
47482320b543307c2d44f3a87a2268ead6faf265c6bd38cf33011e0ac7f8e77a).
This evidence covers the declared Fortemi authority -> React/Core -> HotM
consumer surface and exact Knowledge Shard 2.0.0/full-v1 behavior on those
three platform cells. Windows remains deferred to
Fortemi #1096.
The suite audit in Fortemi #1081 remains NO-GO; this evidence does not prove
suite-wide portability, complete backup, every architecture, launched
GUI/native-dialog behavior, or one shared schema across the static-index,
state-transfer, and live-persistence planes.
What You Get
- Full note CRUD with revision history, soft delete, starring, pinning, and archiving
- PGlite-backed local storage with
opfs,idb, andmemorypersistence modes - Writable canonical record tier that runs without PGlite — journaled IndexedDB records, Bytecask attachment bytes, DB-free shard import/export, and a rebuildable PGlite projection
- Full-text search with PostgreSQL
tsvector/tsquery, phrase search, filters, facets, snippets, bounded metadata predicates, and safe source evidence locators - Hybrid semantic search with pgvector HNSW and BM25 reciprocal-rank fusion
- Virtual embedding-set selectors, cached similarity graphs, and graph/community artifact persistence
- Dynamic and user-authored communities with React graph-controller hooks for source switching
- Tags, collections, inter-note links, SKOS schemes, concepts, and relations
- Knowledge Shard tar.gz import/export with checksums, exact schema/profile negotiation, and receipt-backed component preservation
- Fortemi
source-note-upsert/1.0.0live persistence in PGlite and RecordStore, with bounded atomic batches, checkpoints, replay-safe outcomes, and typed shard loss because source identities are outside current profiles - 10 manifest-backed Fortemi tools plus 11 exported direct tool helper functions
- Optional embeddings, LLM, local-provider discovery, WebGPU detection, and fallback routing
- React 19 hooks for common UI workflows and direct context access for lower-level integration
Runtime Support
| Runtime | Storage | AI capability notes |
|---|---|---|
| Chrome 113+ / Edge 113+ | OPFS recommended | WebGPU available when enabled; Linux Chrome may require flags |
| Firefox 111+ | IndexedDB fallback | WASM embeddings work; WebGPU production support is limited |
| Safari 17+ | Memory or IndexedDB depending context | Use memory for previews and tests when persistence is restricted |
No backend is required. Deploy apps/standalone/dist/ to any static host.
AI Capabilities
fortemi-react supports opt-in capabilities through the runtime capability manager:
| Capability | Runtime | Enables |
|---|---|---|
| Semantic embeddings | transformers.js WASM | Hybrid search, related notes, link discovery |
| Local LLM | WebLLM / compatible provider | AI revision, concept tagging, title generation |
| GPU detection | WebGPU adapter probing | Hardware tier and model-fit guidance |
| Local provider discovery | Ollama, LM Studio, llama.cpp, vLLM, Jan | Remote/local provider fallback routing |
On Linux Chrome, local WebGPU inference commonly needs:
google-chrome --enable-features=Vulkan --enable-unsafe-webgpu http://localhost:5173
Tool Integration
@fortemi/core exports fortemiManifest for bridge registration and direct helper functions for application code. The manifest currently includes:
capture_knowledge, manage_note, search, get_note, list_notes, manage_tags, manage_collections, manage_links, manage_archive, manage_capabilities.
Direct helper exports also include manageAttachments for attachment metadata and blob operations.
Development
pnpm install
pnpm typecheck
pnpm lint
pnpm test:core
pnpm test:e2e
pnpm build
pnpm examples:site:build && pnpm examples:site:e2e # built-gallery smoke test
The repository uses Node.js 22, pnpm 10, TypeScript, Vitest, Playwright, and Vite.
Documentation
| Guide | Description |
|---|---|
| Getting Started | Installation, provider setup, first note, search |
| Package Architecture | Diagram and capability tables for each npm package and how they layer |
| Search | Text, semantic, and hybrid search modes, filters, RRF fusion, snippets |
| Integration Guide | Embedding in React apps, tool helpers, events, jobs, capabilities |
| API Reference | Full API surface for @fortemi/core, @fortemi/graph, and @fortemi/react |
| Deployment | Static hosting, Vite config, browser compatibility, WebGPU, CI/CD |
| Extending | Custom tools, job handlers, capabilities, migrations, hooks |
| Supply Chain | Release signing, workflow pinning, and publishing controls |
License
AGPL-3.0-only. See LICENSE.
fortemi-react MCP questions
How do I add fortemi-react MCP to Claude Code?
Run claude mcp add fortemi-react -- npx -y @fortemi/core in your terminal. Add --scope user to make it available in every project.
How do I add fortemi-react MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
Does this require a backend server?
No, it is a local-first solution that runs the schema entirely in the browser with no backend or account required.
What kind of search capabilities are available?
It supports full-text search using PostgreSQL tsvector, as well as hybrid semantic search using pgvector HNSW and BM25 reciprocal-rank fusion.
Can I use my own AI models?
Yes, you can use local providers like Ollama, LM Studio, or llama.cpp, or route to a provider you control.
How is data stored in the browser?
It uses PGlite-backed local storage with support for OPFS, IndexedDB, and memory persistence modes.