Get ORQVION

Features

Everything that has shipped in the ORQVION desktop platform.

A Tauri IDE, online and offline AI with AUTO routing, a sandboxed coding agent, stack adapters, and enterprise governance — described as implemented, including what is still partial.

Product

Explore what ships in the desktop app.

Dedicated pages for agents, Composer, CLI, models, and team licensing. Online + offline AI in one binary — Ready wizard online-first; your repo is never an ORQVION-hosted coding cloud.

Agents

ORQVION agents plan and execute multi-step work in a Rust ReAct loop on your hardware. They read, search, shell, and edi...

Composer

Composer (Cmd+I) is a dedicated workspace for tasks that span files: a checklist, unified diffs, chat tabs, and checkpoi...

CLI

`orqvion agent` runs the desktop ReAct orchestrator on the machine where you invoke it — same ONLINE · OFFLINE · AUTO ro...

Models

ORQVION ships a master model registry (206 online metadata entries, 49 offline GGUF catalog) with routing modes ONLINE,...

Teams

ORQVION is a desktop IDE with online and offline AI on each seat — not a hosted coding cloud. Business teams get device-...

Help

ORQVION is a desktop IDE with online and offline AI (AUTO routing by default). Teams can self-host a PostgreSQL license...

Environment

A real desktop IDE, not a wrap around a browser tab.

Tauri 2 + Rust + React. Monaco, explorer, command palette, and themes on macOS, Windows, and Linux.

Monaco editor

Syntax highlighting, search, multi-file tabs, command palette, and light/dark themes. The editor is the product — not a sidecar chat box.

Language servers

Autocomplete, hover, go-to-definition, and diagnostics for TypeScript, JavaScript, PHP (Intelephense), HTML, and CSS. Multi-LSP via the adapter registry.

File explorer & preview

Workspace explorer plus an in-app browser preview with live reload for web and Laravel apps. The agent can drive localhost preview over CDP.

Command palette

Keyboard-first actions for files, commands, agent modes, and settings — the same muscle memory developers expect from a serious IDE.

Online & Offline AI

One router — online when connected, offline when you are not.

ORQVION ships ONLINE, OFFLINE, and AUTO routing. Online Beta defaults to online paths (entitlement on the license server + optional BYOK direct to providers). Offline uses ORQVION AI signed bundles and GGUF downloads via llama.cpp — not the Ollama application.

AUTO routing

Task classifier + hardware + network score routes chat, Tab, Cmd+K, Composer, and agents. Manual model pick always wins. Fallback chain: managed/BYOK alternates → verified local GGUF → ORQVION AI bundle.

Online (default)

License server is control plane only — entitlements, not prompt proxy. Desktop calls provider APIs with BYOK keys in your keychain. OpenAI is not a required core provider.

206 online catalog

Master registry metadata for routing and entitlements. Live execution depends on provider uptime, plan, and keys — not every id is a hosted endpoint.

Offline (49 GGUF catalog)

Direct HTTPS downloads with pause/resume (`.part`), ORQVION AI Lite/Standard recommended. llama.cpp inference on your machine. No `ollama serve`, no `:11434`.

Inline completion (Tab)

Editor-native ghost text and multi-line next-edit blocks. Routed by plan, mode, and catalog — online or offline per policy.

Optional BYOK

Bring-your-own provider keys when allowed. Stored in OS keychain / encrypted file — never in logs or telemetry. Org policy can forbid external models.

Agents

Autonomous coding with approvals, diffs, and a sandbox.

A Rust ReAct orchestrator plans and executes multi-step work. Writes are not silent. Docker isolation can be required by policy.

Coding agent

Read, write, search, shell, and Artisan tools. Multi-step tasks with recoverable checkpoints. Default mode has full tools; you still approve writes.

Ask / Plan / Debug / Agent

Ask is read-only. Plan produces a plan you approve before edits. Debug leans on logs, tests, and DAP. Agent is the full loop. Viewers can be locked to Ask.

Composer

A dedicated multi-file session (Cmd+I): task, file checklist, unified diff review, chat tabs, and checkpoints that survive restart in local SQLite.

Diff review

Per-hunk accept or reject for agent and inline edits, plus accept-all. Pending edits use a unified Rust diff model — not a dump into the working tree.

Subagents

Research, Shell, and Review children with isolated context. Custom agents in `.orqvion/agents/*.md`. Default cap of two parallel subagents offline.

Write approval & sandbox

File writes require approval. Agent commands can run inside Docker. Org policy can make the container sandbox mandatory. Secrets never belong in audit logs.

CLI

`orqvion agent "task"` runs the same engine headless — same ONLINE/OFFLINE/AUTO routing as the IDE — for scripts and power users.

Architect path

Architect → Developer → QA mapping on the subagent model for longer builds, with clarification questions before the agent charges ahead.

Context

You control what the model sees.

Mentions, project rules, skills, MCP, and hooks are local files. Org policy can restrict them. ORQVION rules can be imported read-only.

@-mentions

`@file`, `@folder`, `@symbol`, `@codebase`, and `@selection` from a searchable picker backed by the local index — not a cloud embedding upload.

Project & user rules

`.orqvion/rules/*.md` with `alwaysApply`, globs, and description. Global user rules merge in. Org policy can disable or constrain them.

Skills

`.orqvion/skills/**/SKILL.md` discovered by the agent and invokable with slash commands. Sample skills ship in the repo.

MCP

Model Context Protocol client over stdio (offline) with optional SSE. Configure in `.orqvion/mcp.json`. Tools appear in the agent with policy gates and an MCP settings panel.

Hooks

`.orqvion/hooks.json` for PreToolUse, PostToolUse, beforeSubmitPrompt, and stop. A hook can block `file_write` and the block is audited.

Project intelligence

Tantivy BM25, Tree-sitter symbols, SQLite symbol store, and hybrid search. Desktop builds persist vectors in LanceDB (`semantic.lance/`). Online Beta does not require local embedding models to start working.

Workflow

Git, terminal, tests, and debug in the same window.

The agent can run commands, but you keep a first-class IDE surface for the work you still want to do by hand.

Git

Branch, status, diff, commit, and revert via subprocess git. Agent diffs can optionally sync into the Git panel. Attribution/blame tooling is present; GitLab bot parity is not claimed.

Integrated terminal

portable-pty + xterm.js. A real PTY, not a fake output pane. Agent shell tools can reuse the same workspace.

Test runner

Run the project test suite from the IDE. Failures feed Debug mode and self-healing suggestions with guardrails.

DAP debugger

Debug Adapter Protocol sessions for Node and PHP without leaving the workspace. Debug agent mode is wired to logs and DAP, not guesswork.

Studios (R4–R8)

Dedicated studio panels — registry, agents, and honest capability gates.

Media, browser automation, app builder, game/XR, and Web3 each ship a Rust crate, desktop panel, and policy-aware agent tools. They orchestrate your repo; they do not replace Blender, Unity, or a custodial wallet.

Media Studio

Image/video job registry, router, and history in the desktop app. ffmpeg probe/transcode and local generation paths — files stay on disk unless you export them.

Browser & computer use

Policy-gated browser tools for verified navigation and actions. No silent exfiltration; org policy can disable the surface entirely.

Universal App Builder

Framework-agnostic scaffold flows and builder agents for greenfield apps — still subject to write approval and sandbox policy.

Game + XR Studio

Per-engine capability matrix (Unity, Godot, Unreal, WebXR/OpenXR). Panels and `run_game` / `run_xr` assist the workspace; engine editors remain authoritative.

Blockchain + Web3 Studio

Hardhat/Foundry adapters, contract workflows, and security-oriented scans. Mainnet deploy is explicitly gated — testnets and local chains first.

Stacks

One platform from Laravel to XR — adapters, not a single-language demo.

Each stack is a detector, a panel, and agent tools. Studios (above) add depth for media, game, and Web3; nothing here is vapour.

Web

React, Next.js, Vue, Vite and friends: framework detection, LSP, live preview, and a dev-server workflow inside the desktop app.

Laravel / PHP

Project detection, Intelephense, Artisan runner, scaffold wizard, and browser preview with live reload. First-class for Pakistani and PHP teams.

Mobile

Flutter and Expo detection, Mobile panel, and an agent `run` tool for the mobile workspace.

AI / ML

PyTorch and TensorFlow project detection, AI/ML panel, and `run_ml` for local experiment loops — separate from chat LLM routing (online/offline AUTO).

Web3

Hardhat and Foundry detection, Web3 panel, and contract-oriented security scans — not a custodial wallet.

Game

Unity, Godot, and Unreal hooks plus a Game panel and `run_game`. Engine editors stay the engines; ORQVION orchestrates the repo around them.

XR

WebXR / OpenXR panel and `run_xr` for XR project workflows.

Media & video

Stable Diffusion generate/convert, Media panel, ffmpeg transcode/probe, and Video panel. Local files stay local.

Planning

PM planning, clarification questions, Employee panel, and `run_pm` so a brief becomes a sequenced build instead of a single prompt dump.

Unified workspace (R10)

One nav for IDE, agents, and enterprise collaboration.

Unified workspace navigation groups Studios, collaboration, and core IDE sections without duplicating Private Team Mesh. Collaboration metadata syncs; project source stays local by default.

Unified section nav

Desktop `UnifiedWorkspaceNav` highlights the active section only — Agents are not falsely “always on” when you are in the IDE.

Enterprise collaboration

Teams, activity feeds, and enterprise controls on the collaboration crate — mesh chat/share preserved alongside governed org features.

Privacy defaults

No server-side file storage for your tree by default. Audit covers security-relevant actions, not prompts or file contents.

Operations

Health, extensions, updates, and licensing on the device.

Self-healing is guarded. Marketplace packages are signed. License keys are device-bound.

Self-healing

Health diagnostics and safe repairs for the app and, with guardrails, the project. GPU/LSP heal metrics exist on the enterprise path. The agent cannot silently rewrite your tree.

Signed marketplace

Extension catalog with signed packages, a WASM sandbox runtime, and a model CDN. Unsigned or policy-blocked packages do not run.

Updates

Desktop updater with Ed25519-signed entitlements. You choose when to apply. Offline allowances exist per plan.

Licensing

Free / Pro / Business / Enterprise. Device-bound seats. PostgreSQL license server for orgs. JazzCash, Easypaisa, Alpha, and bank transfer on the same catalog as desktop.

Enterprise

Governed agents for regulated teams — inside your perimeter.

SSO, RBAC, org policy, central audit, and a self-hosted license server. Source code is not the payload we store.

Identity

OIDC, SAML 2.0, and LDAP / Active Directory. Org roles: Owner, Admin, Developer, Viewer. Desktop sessions refresh against your IdP.

Policy engine

Disable external models, require containers, scope tools, restrict rules/skills/MCP, and lock viewers to Ask mode.

Audit

Structured JSON events, sync-ready for the enterprise server. Security-relevant actions only — not file contents, prompts, or secrets.

Secret scanning

Hooks on the enterprise path to catch credentials before they land in a diff you were about to apply.

On-prem license server

PostgreSQL-backed server, Docker Compose, and Kubernetes HA options. Private cloud and air-gapped designs are in the enterprise conversation — not a public SaaS tenant you must join.

Private Team Mesh

Desktop peer mesh for encrypted team chat, channels, file share, and relay (M0–M9). Works alongside R10 collaboration controls — not a replacement for org SSO or audit. Mobile protocol validation (M10) is next — not claimed as a polished mobile mesh yet.

Pakistan billing

PKR packages via JazzCash, Easypaisa, Alpha, and bank transfer. International plans exist; you are not forced through a US-only card processor.

Honesty

Shipped does not mean every research box is closed.

These limits are documented in the engineering roadmap. We would rather say them on the Features page than overclaim in a hero.

  • LanceDB semantic search ships in the desktop app (`lancedb` feature). CLI and server builds without that feature still use SQLite + in-memory cache.
  • ONNX Runtime ships behind the `onnx` feature with a pinned CPU CI benchmark on a mini embed fixture. Production embed paths do not call an Ollama daemon.
  • GitHub review-bot work is behind a feature flag. GitLab bot parity is not shipped.
  • Cloud background agents on ORQVION-hosted GPUs are out of scope. Online uses BYOK/direct providers; offline uses ORQVION AI + GGUF on your hardware.
  • Private Team Mesh ships M0–M9 on desktop (P2P E2E chat, channels, share, relay). M10 mobile protocol validation is still planned — do not treat mobile mesh as GA polish.
  • Game, XR, media, and ML panels are MVP adapters: they detect, run, and assist. They do not replace Unity, Blender, or a full MLOps suite.
  • Media Studio transcode/generation depends on local ffmpeg and GPU paths you enable — we do not claim cloud render farm parity.
  • Browser/computer-use agent tools require explicit policy; they are not an always-on remote desktop.
  • Blockchain Studio mainnet deploy is policy-gated; testnets and local chains are the default safe path.
  • R10 collaboration syncs team metadata — not your repository. Unified workspace nav is desktop UI; it is not a second cloud IDE.
  • Customer portal: Free signup at /portal/signup or magic-link at /portal/login — orders and support metadata, not desktop SSO.
  • Online Beta Ready wizard defaults to online — Advanced Local (offline GGUF / ORQVION AI) lives under Settings and is optional at first launch.
  • Legacy catalog ids may contain “ollama” in the name — they map to HuggingFace GGUF downloads, not the Ollama app.

Why not a cloud AI IDE

Same job. Different data boundary.

Topic Typical cloud AI IDE ORQVION
Source code default Sent to a vendor cloud for the assistant to work Stays on your machine. No permanent ORQVION storage of prompts or code.
Default AI path Vendor-hosted assistant that requires their cloud Online + Offline AI (AUTO). Default online; ORQVION AI + GGUF offline — no Ollama app.
Agent execution Often unsandboxed shell on the laptop Write approval plus optional Docker container sandbox.
Enterprise control You join their SaaS tenant SSO, RBAC, org policy, and a self-hosted license server.
Pakistan billing International cards, USD pricing JazzCash, Easypaisa, Alpha, and bank transfer — same catalog as desktop.

ORQVION uses the network for licenses, entitlements, updates, and payment. Online inference goes to providers you enable (BYOK); offline stays on-device (ORQVION AI + GGUF). Never a silent dump of your repository.

Read the controls, then install the desktop app.

Docs cover online/offline setup, AUTO routing, and org rollout. Legal pages cover privacy, AI/data processing, acceptable use, terms, refunds, and returns.

Get ORQVION Docs