TARX is a governed runtime for personal and enterprise AI. It gives models and agents the operating layer they are missing: persistent context, private memory, approved capabilities, model routing, human approval, and evidence of what happened.
Computer by default. Supercomputer by permission.
Website · Download the Mac beta · Documentation · Public releases
flowchart LR
U["Person or enterprise workflow"] --> X["TARX"]
X --> M["Memory and context"]
X --> T["Tools and agents"]
X --> G["Permissions and approvals"]
X --> E["Evidence and audit"]
X --> R{"Approved route"}
R --> L["Local Computer"]
R --> P["Private infrastructure"]
R --> C["Approved cloud models"]
TARX is not a model company or a chatbot wrapper. It is the governed layer around inference and action, designed so a user or organization can control where work runs, what an agent may do, and what proof it must return.
| Repository | What you can inspect |
|---|---|
tarx-desktop |
Computer-canonical Mac beta (app.tarx.com/computer), signed/notarized releases, navigation + agentic smoke, resource bounds |
tarx-cli |
tarx route check local fail-closed preflight; four-host MCP install fixtures (Claude Desktop/Code, Cursor, VS Code) |
governed-agent-contracts |
Proposal → decision → result → evidence with executable invariants; MCP + enterprise route fixtures |
| Founder: Palantir AIP × local-first | Independent enterprise reference: policy routing → TARX computer / private / deny (not affiliated with Palantir) |
The current public surface demonstrates:
- Computer by default product entry (not a marketing chat wrapper)
- Fail-closed local and approved remote inference preflight
- MCP interoperability with host-adapter fixtures and merge invariants
- Proposal-first actions; Supercomputer only with explicit approval
- Enterprise local-first routing and offline/policy-deny evidence
- Health, system-integrity, release, and security documentation
The best public contributions improve the developer boundary: compatibility, diagnostics, release trust, MCP interoperability, documentation, and reproducible tests. Proprietary orchestration, internal operations, production infrastructure, customer data, and model weights remain private.
Start with the TARX CLI or install the TARX Desktop beta.
TARX is built by TARXAN Inc, founded by John Wantz Jr., an AI systems architect and product leader focused on governed agents, local-first AI, enterprise search, evaluation, and human-centered AI systems.