Model Management
Run any model, anywhere — directly from the IDE.
Neural Inverse ships a full model management layer inside the IDE. Install base models with one click, deploy private GPU endpoints to your own AWS or Azure account, and see the live status of every model in your environment — all without opening a terminal or leaving the editor.
Quick Install
Open Agent Manager (Cmd+Alt+A) → Models tab. Pick a model, click Install. That's it.
Neural Inverse detects your local Ollama instance automatically and streams the download directly into the card — no terminal, no manual ollama pull.
| Category | Available models |
|---|---|
| Code | Qwen2.5-Coder 7B / 14B / 32B, DeepSeek-Coder V2, CodeLlama 13B / 34B, Codestral |
| Chat | Llama 3.3 70B, Llama 3.2 3B, Phi-4 14B, Gemma 3 12B, Mistral 7B |
| Reasoning | DeepSeek-R1 7B / 14B / 32B, QwQ 32B |
| Multimodal | LLaVA 13B, Llava-Phi-3, BakLLaVA |
Once downloaded, the model is auto-detected and registered in your provider settings. No manual configuration.
Cloud Deployment
Deploy a private vLLM endpoint to your own AWS or Azure account in a few clicks. Your data never touches Neural Inverse infrastructure.
What Neural Inverse provisions for you:
- GPU instance (your choice of type and region)
- IP-restricted security group scoped to your IP only
- vLLM installed and started as a systemd service with auto-restart
- A generated API key so your endpoint is never open
- IMDSv2 enforced on EC2 (no metadata SSRF)
The wizard shows a live provisioning timeline — every step as it happens, elapsed time, and an Abort button if you need to cancel. Deployments that get stuck are automatically recovered.
Deployments Tab
A live view of everything running in your model environment.
- Local providers — Ollama, vLLM, LM Studio detected automatically via health checks every 30 seconds
- Cloud deployments — all your AWS/Azure GPU instances with status, cost/hour, and quick actions
- Live updates — status changes appear in real time without manual refresh
Auto-Configuration
When Neural Inverse detects a new deployment, it automatically configures the matching provider in your LLM settings — but only if you haven't already set it up.
- Skips configuration if you've set a custom endpoint, API key, or models list
- Shows a notification with OK, Undo, and Don't auto-configure options
- Dismissal is remembered permanently per provider
Enterprise
Neural Inverse Enterprise adds organization-wide model governance on top of the open source layer:
Neural Inverse is an open source AI-native IDE for critical software:
- 117 GitHub stars
- 357+ MCU variants supported
- 36 migration language pairs
- Apache 2.0. Free and open source. No credit card required.
Private cloud model registry
Deploy models once to your org's private cloud or network and share the endpoint across every developer on the team. No one needs to manage their own instance.
- Single deployment, team-wide access
- Endpoint credentials managed centrally — developers never handle raw cloud keys
- Works within your existing VPC / private network — no public internet exposure required
Policy enforcement
Define which models and providers developers are allowed to use. Unapproved providers are blocked at the IDE level.
- Allowlist models by provider, parameter count, or capability type
- Block cloud providers not approved by your infosec team
- Policies propagate to all clients within 30 seconds of a change
Local model governance
IT defines an approved local model catalog. Developers install from the approved list only — no arbitrary ollama pull from the internet.
- Curated catalog visible to developers; unapproved models hidden
- Audit log of every model install and usage across the org
- Works with Ollama, vLLM, and LM Studio
Centralized credential management
Cloud credentials (AWS, Azure) are stored and rotated at the org level. Developers connect to shared deployments without ever seeing raw access keys.
Usage telemetry
- Token counts per developer and per project
- Model usage breakdown by team
- Cost attribution per deployment
SSO-gated deployment access
Only authorized roles can provision or terminate cloud GPU instances. Enforce via your existing SSO provider.
FAQ
Does this require a Neural Inverse account?
No. Quick Install and cloud deployment work entirely with your own Ollama install and your own AWS/Azure account. Neural Inverse does not proxy any traffic.
What cloud providers are supported for GPU deployment?
AWS EC2 and Azure VM today. GCP is on the roadmap.
What happens if provisioning fails mid-way?
The wizard shows a live error log. You can retry or abort. Any deployment stuck in provisioning for more than 20 minutes is automatically transitioned to error state on next IDE start so it doesn't block the UI.
Can I use my own vLLM or Ollama endpoint instead of deploying a new one?
Yes. Neural Inverse auto-detects any Ollama, vLLM, or LM Studio instance running on localhost. For a remote endpoint, add it manually under Settings → LLM Providers.
How does auto-config interact with settings I've already set?
Auto-config only runs if the provider has no custom endpoint, no API key, and _didFillInProviderSettings is false. If you've touched any setting for that provider, auto-config skips it entirely.
What is the Enterprise pricing for model management?
Model governance and the private cloud registry are part of Neural Inverse Enterprise. Talk to sales for pricing.
Related
- LLM Providers (BYOLLM) — connect cloud providers
- Model Management Docs — full reference
- Contributing: Model Management — architecture guide for OSS contributors
- Enterprise — full Enterprise feature set