HashCortx # HashCortx ### Every AI model. One desktop. Nothing leaves without you. Chat, a coding agent, agent teams, a 3D CAD forge and an ERP builder,
on your keys and your machine. No backend, no telemetry, no account. **In the making since 1 May 2026.** Download Website Watch the 3.0 release film Watch HashCoder and Ollama Wiki Discuss on Hacker News Discussions CI MIT macOS Apple Silicon Windows 10 Built with Tauri v2 Version 4.5.0
Hash-7777%2FHashCortX | Trendshift

Nothing phones homeNo analytics, no crash reports, no update checks. Every connection is for work you asked for, and SECURITY.md lists each address.
Every model, side by sideTwenty-one cloud providers, Ollama, and any local model app that speaks the common chat format, with your own keys. Fully local whenever you want, with a local model kept on your graphics chip and loaded between steps.
Your documents, searched on your machineA local knowledge base (RAG): your PDFs and notes are indexed and searched on this computer, by meaning and by keyword, with the search model built into the app. How.
An agent that asks firstCommands, deletions, web pages and anything outside your project wait for your yes. Every file change is a diff you can keep or undo.
Readable to the last lineMIT. Vanilla JavaScript and Rust, no bundler, no minified app code.
> **Version 4.5.0, kept whole and kept private.** HashCoder keeps every conversation whole and never looks through your own folders unless you ask, a source in a reply opens the page it came from, and chat replies are formatted as they are written. [What is new](https://github.com/Hash-7777/HashCortX/releases/tag/v4.5.0). ## See it run
Play the HashCortx 3.0 release film on YouTube: the app mark and the name HashCortx
The 3.0 release film · 0:58
Every model side by side, the harness, the Agent Swarm and each workspace, private by design.
Play the HashCoder and Ollama film on YouTube: the title The HashCoder Harness
HashCoder, the harness and Ollama · 0:52
A 30B model on Ollama finishes only when the tests pass, then a 3B model codes in light mode.
Play the film HashCortx 3.0, every model in one workspace, on YouTube: the name HashCortx 3.0 and the line The AI workspace for everything you make
Every model, one workspace · 1:13
A tour of HashCortx 3.0: ask any model, and compare two side by side as they answer.
Play the HashNotch film on YouTube: the name HashNotch above the line Free and open source
HashNotch, the notch made useful · 0:52
From the Hash ecosystem: your music, your Mac's vital signs and its notices in the MacBook notch.
## Seven workspaces
ChatsEvery provider in one conversation, with projects, attachments, memory, a local knowledge base, a queue for messages sent while a model works, versions of every edited message and regenerated answer, and agents that show what they searched and read, with the source of each sentence marked
HashCoderA coding agent on your real project: file tree, diffs, terminal, Undo
Agent SwarmTeams of agents on any mix of models, built on a canvas or designed from your task
3D ForgeDescribe a part, get a dimensioned solid in its own colours. STL, OBJ, 3MF and STEP, in millimetres
ERPDescribe your business; its agent builds a working app with records, stages and books, exported as one HTML file
FinanceStatements, CSV, PDF and XLSX read into figures, every total and chart worked out by the app
SandboxAgents that inspect untrusted code for malware, prompt injection and suspicious logic
Chats: a question about SQLite journal modes answered by an agent that searched the web and read two pages, a comparison table with a numbered source mark after the sentences each source backs, how many sentences come from what it read, and the two sources it read as cards Plus **Agents**, nine built-in specialists and a builder for your own, **Split**, one prompt answered by two models side by side, and **Connections**: a business system you connect over MCP (Odoo, GitHub, Stripe and Supabase are ready-made), whose key never reaches a page or a model and whose every change asks you first. [How](docs/SECURITY.md#connected-systems). ## HashCoder and its harness It works on your real files. Every edit is a diff with **Keep** and **Undo**, tests run in the built-in terminal, and screenshots, PDFs and files attach by paste or drop. Around the model sits a harness: what the model is shown, what it may do without asking, and what it has to prove before a run may end are decided by HashCortx, not by the model. The HashCoder harness: it reads the project, plans ask by ask, edits under your permission mode, proves the change with your tests, takes a second look at larger changes, and sends itself back until the work is proven. Small models get light mode, mid-sized models focused tools, larger and cloud models the full harness. **Every run is checked by the app, not by the model's word.** It **reads** the project, its notes for agents and its latest commits; **plans** your request ask by ask; **edits** as diffs you keep or undo, under your permission mode; **proves** the change by code alone, running your tests and reading a site the way a browser does, never by opening, serving or taking a picture of what it made; takes a second look at larger changes as a **reviewer**; and **sends itself back** for missed asks, leftovers and unsourced details, until the work is proven. Every model is given an outline of a larger project's folders as a conversation begins, and before a request's first step the files it needs are found and named to it: by the app's own search, and for a model of 5B and up by a reading-only search of its own, whose reads never join the run. A file is changed only once it has been read in the conversation and has not changed since, and a change to a name other files use waits until those files have been read. It is fitted to the model: under 5B, **light mode**, where the model writes whole files and HashCortx does the tool work and runs the tests; 5B to 15B, **focused tools** with the steps written out and a plan for longer requests; 15B and up, and cloud models, **the full harness**, with a map of your code, lessons per project and a reviewer for larger changes. - **Proves its work.** It runs your tests after a change, reads a website the way a browser would before calling it done, and sends itself back for what it left unfinished: a renamed name still written somewhere, files that do not load each other, the example your request gave. - **Fits the model.** Cloud models get a map of your code; smaller local models read one line per tool, and small ones get a light mode that writes whole files while HashCoder does the tool work. - **Does what you asked, and no more.** No features, files or fixes you did not ask for, and an answer of a few plain sentences: what changed, where, and what ran. - **Keeps what you did not ask to change.** An edit that drops a function, a section or a heading is sent back, and what you say to leave alone is left alone. - **Works in your repository.** Its own git tools read the status and the changes, commit when you ask, signed as your commits are, and make a branch, push or pull only after asking you each time, never forced. - **Offers a way on when it stops short.** A run that ends stuck, at its limit or with parts of a several-part request still open offers once more with a hint, or a fresh start on a stronger model you have, from your request and what it already did. - **Hands off part of a big task, if you switch it on.** New, and not yet proven on real runs, so it is off until you turn it on: a strong model can start sub-agents on the same model, and any model at full size can when your request asks for them, one that reads and reports or one that changes only the files it is given, and HashCortX checks each report before the model reads it. - **Keeps out of your own folders.** It does not list or search your home folder, Desktop, Documents or Downloads unless your request asks it to look there, and a folder it makes for your request becomes the project. - **Asks as much as you choose.** Manual asks before every change and is where it starts; Accept edits writes project files freely; Auto adds a fixed list of read, search and check commands. HashCoder after a run on a scripted stand-in model: the request to make npm test pass, the run signal across its checking, reading, changing and checking phases, the answer, the edit to range.js with Keep and Undo, and the receipt saying npm test passed ## Your documents: a local knowledge base (RAG) HashCortx answers from your own files without sending them anywhere to be indexed. Retrieval-augmented generation (RAG) runs entirely on your computer. - **Add documents by attaching them.** With the knowledge base switched on (in Agents), a PDF or text file attached in chat is split into overlapping passages of about 1,200 characters and stored on this computer, in the app's own storage. - **Every passage is indexed twice:** by its keywords, and as a 384-number meaning vector made by `bge-small-en-v1.5` (MIT, BAAI). That search model is compiled into the app and runs in its Rust core through ONNX Runtime: nothing is downloaded and nothing is uploaded to index or search. - **Search by meaning and by exact words.** A question is ranked against both indexes and the two rankings are fused (reciprocal rank fusion), so "chief executive" finds a passage about the CEO, and an error code or a rare name is still found exactly. The best few passages are used, not whole documents. - **Where it is used.** Chat's agents look things up with a `search_knowledge` tool, and so does HashCoder on larger local and cloud models; both are told to cite the source they used. In chat, `/inject` adds the best passages to a message sent to a local model. The passages become part of the request to the model you chose, so with Ollama they never leave your machine. - **Works offline,** on the first launch, with no setup. The Windows build leaves the search model out so it starts on any 64-bit PC, and searches by keyword only. - **What else goes in.** Results of the agents' web, Wikipedia and PubMed searches and pages they read are also kept in it, labelled with where they came from. How it is kept private, and what it is not, is in [SECURITY.md](docs/SECURITY.md#the-knowledge-base). ## Agent Swarm Build a team on a canvas, start from a template, or describe the task and have one designed, sized to the job. Watch every agent in the live trace, then open the result: each agent's part, the files it made and every version of them. Agents that did not finish can be run again on their own. A website a team builds can show real, openly licensed photographs, credited on the page. Agent Swarm: a four-agent team mid-run on a canvas, with the live trace below
3D Forge: a desk lamp generated as four parts, 69 by 100 by 69 millimetres ERP: a guesthouse system built by its agent, its reservations by status, with the agent's conversation open beside it
3D Forge. A described object becomes real parts with real dimensions, ready to print or open in CAD. ERP. Tell the agent about your business and it builds the app. Ask it to change the app, record what happened, or answer from your records.
## Security, stated plainly HashCoder asking permission to run npm test in the project, with Deny, Allow for session and Allow once - **You approve what matters**, and every decision is written to `~/.hashcortx/audit.log`. - **A blocklist compiled into Rust** refuses keys, credentials and system folders, as files or inside commands. - **On macOS, agent commands run in the system sandbox**, kept out of your keys, keychains, shell start-up files and the app's own data. - **A task on a local model stays local**, and cloud requests are capped at 30 a minute and 6 at once. - **What it is not:** keys are stored unencrypted, protected by your user account, and the build is not code-signed. Everything it does, and does not, is in [SECURITY.md](docs/SECURITY.md). ## Install **macOS (Apple Silicon).** Download the DMG from the [latest release](https://github.com/Hash-7777/HashCortX/releases/latest) and drag HashCortx to Applications. The build is not notarised: on first launch, right-click the app, choose **Open**, then **Open** again. If macOS still refuses: ```bash xattr -dr com.apple.quarantine /Applications/HashCortx.app ``` **Windows.** Run the installer from the [latest release](https://github.com/Hash-7777/HashCortX/releases/latest). It is built without the embedding model, so it starts on any 64-bit PC and searches the knowledge base by keyword. **Linux and Intel Macs.** Build from source, below. Then open **Settings → API keys**, add a key and press **Test**, or run a model on your own machine with **Settings → Local model**.
Build from source
```bash cd ~ git clone https://github.com/Hash-7777/HashCortX.git cd HashCortX npm install npm run tauri dev # develop npm run tauri build # package ``` Needs Node 18+ and Rust via `rustup`, plus **macOS:** Xcode Command Line Tools · **Linux:** Ubuntu 24.04+ with the Tauri v2 system libraries · **Windows:** MSVC build tools and WebView2. **On Windows, run `cd ~` first.** An administrator PowerShell starts in `C:\Windows\System32`, and a checkout there fails to bundle with a misleading "file not found". **On Windows, allow the app in Ollama.** The app's page comes from `http://tauri.localhost`, which Ollama refuses until it is allowed. Run `setx OLLAMA_ORIGINS "http://tauri.localhost"` and reopen Ollama; until then the app says that the local model app refused it. **Older x86-64 processors (without AVX2).** The default build links an ONNX Runtime that needs AVX2 and BMI2 (Intel Haswell, AMD Excavator and newer); an app built with it closes before a window appears on an older processor. Building on such a processor stops with a message that says so, early in the build, before the app itself is compiled. Build without the embedding model instead, and search falls back to keywords: ```bash npx tauri build -- --no-default-features npm run tauri dev -- -- --no-default-features # and to develop cargo test --manifest-path src-tauri/Cargo.toml --no-default-features # and to test ``` Before a pull request, run what CI runs: every source check on Linux, the ones that could depend on the system (`npm run check:windows`) on Windows too, and the Rust tests on all three: ```bash npm run check # 13,166 source checks cargo test --manifest-path src-tauri/Cargo.toml # 267 Rust tests ``` Tauri v2 · Rust · vanilla JavaScript with no bundler, served as written. [Architecture](docs/ARCHITECTURE.md) · [Security](docs/SECURITY.md) · [Contributing](CONTRIBUTING.md) · [Changelog](CHANGELOG.md)
## FAQ **Is it free?** Yes. MIT, no paid tier. You pay your AI providers directly, or nothing with a local model. **Does it work offline?** Yes, with Ollama or another local model app. Web search, cloud models and the Python sandbox's first start need the internet. **Which systems?** macOS on Apple Silicon, built and used daily. Windows 10, tested. Linux builds and passes CI, but has not been run by anyone yet. **Does my code leave my machine?** Only inside a request to the model you chose. There is no HashCortx server. **Are my keys encrypted?** No. They sit in the app's own folder, protected by your user account. [Why](docs/SECURITY.md#where-api-keys-live). More in the [Wiki](https://github.com/Hash-7777/HashCortX/wiki/FAQ). | Shortcut | Action | |---|---| | `Cmd/Ctrl + Shift + C` | Open or leave HashCoder | | `Cmd/Ctrl + Shift + N` | New chat | | `Cmd/Ctrl + K` | Model picker | ## The Hash ecosystem | App | What it is | Licence | |---|---|---| | **HashCortx** *(you are here)* | The local-first AI workspace | MIT | | [**HashCerebrum**](https://github.com/Hash-7777/HashCerebrum) | Medical research workbench with a 3D brain interface | AGPL-3.0 | | [**HashMeterAi**](https://github.com/Hash-7777/HashMeterAi) | An honest local meter for your AI usage | Apache-2.0 | | [**HashNotch**](https://github.com/Hash-7777/HashNotch) | Turns the MacBook notch into a live activity island | GPL-3.0 | They connect through files on your disk, not a service: HashCortx records token counts in `~/.hashcortx/usage.jsonl` for **HashMeterAi**, and posts a short "finished" notice, a title and never a prompt or an answer, for **HashNotch**.
**HashCortx** · since 1 May 2026 · by [Seif Hashish](https://github.com/Hash-7777) · [MIT](LICENSE) [Download](https://github.com/Hash-7777/HashCortX/releases/latest) · [Website](https://hashcortx.com) · [Wiki](https://github.com/Hash-7777/HashCortX/wiki) · [Discussions](https://github.com/Hash-7777/HashCortX/discussions)