Install NeuraFrame Studio™ on ARM64 or x86_64 Linux. Start with a 7-day free trial per device. Not sure what you need or which to choose? Start here: the three answers and a 10-minute first run.
For Jetson Orin, ARM64 edge systems, robotics platforms, and supported ARM64 Linux devices.
SHA256: verify checksum
For Linux workstations, servers, local LLM hosts, and development systems.
SHA256: verify checksum
Do not replace your model provider. Put NeuraFrame™ in front of it. The x86_64 build runs a reuse-aware proxy in front of your provider: point your app's base URL at NeuraFrame™ and change nothing else. Repeats are served from memory; everything else is forwarded to your provider unchanged. If NeuraFrame™ has no stored answer and cannot reach your provider, it says so plainly. It never makes one up.
For applications already using OpenAI-style chat, completion, or compatible API patterns.
For applications routing Anthropic-style model requests through a server-side AI gateway.
For teams using Grok or xAI-style model access through configurable API routing.
For private, hosted, or local model endpoints that expose compatible HTTP model APIs.
Download x86_64 (Gateway) Gateway guide
Same x86_64 package, same per-device licensing. The gateway runs on 127.0.0.1:8081; set your provider as the upstream and point your app at it. See the gateway guide.
Run the Fleet server in your own cloud or on-prem to manage curated round-trip learning across a fleet. Devices learn, you approve what should spread, and it is distributed back, signed and non-destructive, with an inventor vault and provisioned device identities. How Fleet works.
Pure Python (standard library only, Python 3.8 or newer), runs on any modern x86_64 or ARM64 Linux. No compiled engine and no NVIDIA requirement. The fleet client already ships in the Studio package, so your devices can join.
SHA256: verify checksum
Studio Fleet shares verified reuse memory across servers and devices: they learn, you approve what should spread, and it is distributed back, signed.
These apply to the Studio and Gateway builds, which share the compiled engine. The Fleet server is pure Python and lighter; see the note below the table.
| Operating system | 64-bit Linux, ARM64 (aarch64) or x86_64. Ubuntu 22.04 LTS or a compatible distribution, with glibc 2.35 or newer, Linux kernel 5.15 or newer, and systemd. |
| Python | The exact CPython your build was compiled for. The engine is ABI-locked, and the installer preflight reads the requirement from the build itself and stops with a clear message on a mismatch. Current builds: x86_64 (server): CPython 3.12 (Ubuntu 24.04's default); ARM64 (Jetson): CPython 3.10 (Ubuntu 22.04's default). Match your OS to your build and no Python work is needed. |
| Memory | 8 GB RAM minimum (16 GB for the 8B-model tier), enforced by the preflight. The floors are sized for running a model beside NeuraFrame™; the reuse engine itself is about 16 MB and the full service idles around 70 MB, so fronting only a hosted provider the floor is conservative. |
| Model | A local model or compatible workflow, or a hosted provider behind the Gateway. |
| Accelerator | NVIDIA GPU optional. Jetson Orin is the proven tier; with no NVIDIA GPU it runs CPU-only on the fallback tier. The engine has no NVIDIA or CUDA dependency, so other ARM64 boards such as the Raspberry Pi 5 can run it as an unproven tier. |
| Activation | Internet for online trial and license activation. Offline licensing available by request. |
Fleet server is the exception: pure Python with the standard library only, Python 3.8 or newer, on any modern x86_64 or ARM64 Linux. It has no compiled engine, no NVIDIA requirement, and is not tied to the exact-CPython rule above.
Install NeuraFrame Studio™ across many machines at once and license the whole fleet from one place, with a single enrollment token.
A download should not quietly run for years while newer releases ship. Every install checks whether it is the current release and tells you, and you decide whether to upgrade. The upgrade is in place and keeps your learned memory, config, and license. Nothing is downloaded or changed unless you ask, and every upgrade is verified against a published checksum, and an Ed25519 signature when release signing is enabled, so a tampered file is refused.
neuraframe update # is this the current release? neuraframe update --apply # download, verify, and upgrade in place (keeps your data)
On a fleet, the admin can see which version each device runs and set a target release; each device still chooses when to move. The check is a plain version lookup and can be turned off in config.
Extract the package, run the installer, and start the trial.
# ARM64 shown; use the x86_64 package on x86 systems # optional: verify the download against its published checksum sha256sum -c neuraframe-studio-jetson-0.6.2.tar.gz.sha256 tar -xzf neuraframe-studio-jetson-0.6.2.tar.gz cd neuraframe-studio sudo bash install.sh neuraframe trial start neuraframe status