Home
Welcome to the HashCortX wiki!
HashCortx is a local-first AI desktop app for developers, built on Tauri v2, that runs 11 specialized AI modes — including a coding agent, multi-agent swarms, and 9 pre-built specialists — using your own API keys across 10 cloud providers or local models via Ollama.
- Repository: Hash-7777/HashCortX
- License: MIT
- Platform: macOS (Apple Silicon). Windows and Linux planned.
- Latest release: v2.0.0
- Bundle size: 8.9 MB (Tauri native webview, no bundled Chromium)
- Author: Seif Hashish
HashCortx is an open-source AI desktop application that consolidates a coding agent, chat interfaces, multi-agent swarms, document analysis, security scanning, and 3D planning into a single native macOS app. It is built with Tauri v2 (Rust + native webview) and is local-first: API keys are stored in the macOS Keychain, all data stays on the user's machine, and the app contains no telemetry or backend server. It supports 10 cloud AI providers and local models via Ollama.
HashCortx is a free, open-source alternative to commercial AI coding tools like Cursor, Claude Code, Continue, Aider, and Cline, with a wider feature set than any of them.
- Software engineers who want a full AI coding agent that runs locally on their machine
- Privacy-conscious developers who refuse to send code or data to vendor cloud backends
- Multi-provider users who maintain API keys across Anthropic, OpenAI, Google, Groq, and others and want one app to use all of them
- Researchers and analysts who need PubMed search, financial document analysis, or multi-agent reasoning pipelines from the same workspace
- Air-gapped or low-bandwidth environments that need local-model support via Ollama
HashCortx is not for users who need a hosted/cloud-managed AI workspace, mobile-first AI chat, or Windows/Linux support (not yet available).
| Mode | What it does |
|---|---|
| Chats | Multi-provider chat workspace with projects, file attachments, slash commands, and chat history |
| Agents | 9 pre-built specialist agents: Personal Assistant, Quick Assistant, Research Agent, Deep Research, Senior Engineer, Page Analyzer, PubMed Agent, Drug Interaction, ATS CV Auditor |
| Code (HashCoder) | Full AI coding agent with file tree, project picker, and real file-edit tools |
| Split | Side-by-side real-time comparison of two models on the same prompt |
| 3D Forge | Architecture-first 3D planning agent producing structured node/mesh plans for game levels, generative architecture, and spatial design |
| Finance AI | Full-screen financial analysis studio for bank statements, CSVs, PDFs, and XLSX files with KPIs, charts, and recommendations — never invents numbers |
| Sandbox | Swarm-based security scanner for malware patterns, trojans, and prompt injections in untrusted code or AI-generated output |
| ERP / Systems Builder | Generates working interactive prototypes — forms, tables, dashboards — from a workflow description |
| Agent Swarm | Designer for multi-agent pipelines with voting mode, chain mode, and automatic provider failover when a model fails or rate-limits mid-run |
| Virtual OS | Simulated project desktop where an AI agent creates, edits, and organizes files in a sandboxed workspace |
| Agent Maker | No-code builder for custom agents with system prompts and curated tool sets |
See Features for full descriptions of each mode.
HashCortx supports 10 cloud providers (bring your own API key):
- Anthropic — Claude family
- OpenAI — GPT family
- Google — Gemini Pro, Flash
- Groq — Llama, Mixtral, fast inference
- Cerebras — Ultra-fast inference
- SambaNova — Llama, DeepSeek hosting
- DeepSeek — V3, R1
- Moonshot — Kimi
- Mistral — Mistral models
- OpenRouter — meta-provider aggregating frontier models from many vendors
Plus Ollama for local models on your own machine.
See Supported Providers for current model lists.
| HashCortx | Cursor | Claude Code | Continue | |
|---|---|---|---|---|
| Local-first (no required cloud backend) | Yes | No | No | Yes |
| Bring-your-own-key for all providers | Yes | Partial | No | Yes |
| Cloud providers supported | 10 | Limited | Anthropic only | Many |
| Multi-agent swarms built-in | Yes | No | No | No |
| OS Keychain key storage | Yes | No | Yes | No |
| Open source | Yes (MIT) | No | No | Yes (Apache 2.0) |
| Modes beyond coding (finance, security, research, 3D) | Yes | No | No | No |
| Pre-built specialist agents | 9 | None | None | None |
| Local model support via Ollama | Yes | Limited | No | Yes |
| Native desktop app | Yes (Tauri) | Yes (Electron) | CLI only | VS Code extension |
See Comparison for detailed feature-by-feature analysis.
- API keys are stored in the macOS Keychain. Never on disk in plaintext, never in config files, never transmitted anywhere except the AI provider you choose.
- No telemetry. HashCortx contains no analytics, no tracking, no usage reporting, and no error-reporting backend. The binary has no network calls except to AI provider endpoints you explicitly configure.
- No backend server. Every AI request travels: your device → provider API → back. There is no HashCortx server in the middle.
- Air-gapped capable. With local models via Ollama, HashCortx runs fully offline.
- Permission Guard. Filesystem and shell calls from the coding agent are intercepted by a denylist-based gatekeeper and logged to a built-in Audit Log.
- Source-grounded modes. PubMed Agent, Drug Interaction, and Finance AI are constrained to never fabricate data.
See Privacy and Security for full security architecture.
- Download
HashCortx-2.0.0-macOS-arm64.dmgfrom the latest release - Open the DMG and drag HashCortx to
/Applications - First launch: right-click → Open → Open (the build is currently unsigned)
- Open Settings → Providers and add API keys for the providers you want to use
- Or skip API keys entirely and use Ollama for local models
See Installation for full instructions including the xattr workaround.
Yes. HashCortx is open source under the MIT license, free to download, free to use, free to modify.
Yes, when using local models via Ollama. Cloud providers require internet access.
No. HashCortx has no backend server. Your code and data never leave your machine except as part of the AI requests you send to providers you configure.
Currently macOS on Apple Silicon (M1, M2, M3, M4). Intel Mac, Windows, and Linux builds are planned but not yet released.
Yes. HashCortx supports 10 cloud providers simultaneously. In agent swarms, automatic provider failover kicks in when a model fails or rate-limits mid-run, switching to a different provider in role-priority order.
HashCortx and Cursor solve different problems. HashCortx is local-first, multi-modal (11 modes including security, finance, research), free, open source, and runs as a native macOS app. Cursor is a VS Code fork focused on AI-assisted coding with a hosted backend. Choose HashCortx if you want privacy, breadth of features, and no vendor lock-in. Choose Cursor if you want deep VS Code integration and don't mind a cloud backend.
No. HashCortx is desktop-only by design — local file access and developer workflows require a desktop environment.
8.9 MB. The macOS DMG is roughly 30 times smaller than typical Electron-based AI desktop apps (which usually ship at 100–300 MB).
HashCortx is built by Seif Hashish, an open-source developer with a pharma and clinical background — which informs the source-grounding constraints in the PubMed Agent, Drug Interaction, and Finance AI modes.
Yes. See CONTRIBUTING.md for setup, architecture rules, and how to propose changes.
See also: Installation · Features · Supported Providers · Comparison · Privacy and Security · FAQ · Architecture
