Architecture, specs, and developer surface
Technical depth lives here—how installs work, what ships in the stack, and how VSCodium-Rust differs from generic AI IDEs. Context-heavy work is backed by KORTEX (AGPL-3.0 neural VFS + runtime kernel): parametric paging for 8GB-class local GPUs and relaxed cloud-burst paths on pooled large-VRAM hosts. New here? Start with the homepage or pricing.
From install to inference in three steps
Install Community (free)
Clone VSCodium-Rust from GitHub, build or use release binaries, open your repo. The iPhone-class emulator consumes your local CPU/GPU—you operate fully on-prem if you want.
Attach models
Run local Ollama, subscribe to Cloud Inference for managed Ollama capacity, or wire BYOK providers. Mix and match per workspace—no single mandatory cloud.
Iterate with agents
Point agents at the emulator surface and source tree. KORTEX keeps repo context in a neural VFS with JIT paging—so you are not limited to bruteforcing whole trees into the context window—whether you run local Ollama, Cloud Inference, or BYOK (e.g. Qwen 3.6 35B abliterated, Gemma 4).
Technical building blocks
Editor, guest surface, KORTEX context substrate, agents, and optional pooled Cloud Inference—open AGPL components where noted, proprietary emulator where licensed.
VSCodium-Rust IDE
Rust/Tauri shell, VS Code–compatible, telemetry-off defaults. Lightweight vs Electron IDEs.
Integrated iOS emulator
Proprietary hypervisor guest—iPhone emulator in the sidebar on Windows & Linux. Not Apple's Xcode Simulator. Closed-source; ships with the license.
KORTEX · runtime kernel
AGPL-3.0 neural VFS for sovereign inference: compact semantic map (“limbic” gist), attention-gated JIT page-in, NVMe→VRAM-friendly streaming, LRU eviction under a fixed VRAM budget, and thermal-aware throttling on local AMD-class GPUs (e.g. 8GB profiles). Runtime bridge adapts llama.cpp; cloud-burst mode relaxes strict paging on large-VRAM pooled hosts so shared Cloud Inference stays economical—not one GPU per subscriber.
Agent runtime
Vision + code paths via Ollama and managed endpoints: Qwen 3.6 35B abliterated, Gemma 4, OSS catalog—fed by KORTEX-structured context instead of dumping whole repos into the window.
Cloud Inference (optional)
Subscription add-on: managed Ollama endpoints and curated open-weight catalog—when you want inference hosted for you instead of operating it yourself.
BYOK & voice
Bring your own keys for any HTTP model. Voice-first interactions optional.
Why teams pick this over generic AI IDEs
A product surface that respects both executives and ICs: local-first emulator, optional managed inference, and flat monthly tiers on the pricing page.
Typical stack vs VSCodium-Rust
| Capability | Usual | Here |
|---|---|---|
| What you ship | IDE-only or emulator-only | IDE + emulator bundle |
| Runtime | Electron | Rust / Tauri |
| Codebase context | Huge pastes / window stuffing | KORTEX neural VFS · JIT paging |
| AI path | Single vendor cloud | Local · Cloud Inference · BYOK |
| Mobile target | Mac-centric | Windows & GNU/Linux |
qwen3.6 · 35b-abliterated · gemma4 · ollama
KORTEX paging
AGPL stack: semantic gist, attention-gated inflation, and VRAM-budget eviction—local 8GB-friendly, cloud-burst when large VRAM backs the pool.
Linux-native
Parrot-grade workflows: frida, radare2, kernel observability without leaving the tree.
Vision on glass
Agents read the emulator buffer for pixel-accurate triage—not just logs.
Model routing
Switch Cloud Inference, on-box Ollama, or your keys per workspace.
Three pillars, one workstation
Editor + emulator + inference you control—Community runs the emulator locally; paid tiers add cloud capacity when you want it.
VSCodium-Rust Workstation
The product you license: Rust/Tauri IDE + integrated proprietary iPhone emulator as a single install. VS Code ecosystem compatibility, voice hooks—and KORTEX (AGPL-3.0) for infinite-fidelity-style neural VFS / JIT context, paired with an iOS-class hypervisor guest on non-Mac iron—not Apple's macOS-only Simulator.
Agent Runtime · Ollama · optional cloud
Agents route through Ollama: flagship pulls like Qwen 3.6 35B (abliterated) and Gemma 4, plus the rest of the open catalog. Subscribe to Cloud Inference for managed endpoints, or run local Ollama and BYOK only—see pricing for Pro, Security Researcher, and Enterprise tiers.
Integrated iPhone Emulator
STYX-backed hypervisor guest beside Android lanes—pixels, console, and hooks in one IDE surface. Full device emulation for pentest, malware analysis, and mobile engineering on Windows & Linux—without open-sourcing ACHERON.