Parta Runtime
Local inference · Evaluation to governed production
Run local GGUF inference without assembling a Python or container stack. Start with a free portable build, then move to production where requests can be verified before compute, execution context controlled, and defined events recorded with identity-backed provenance.
Scroll
Less infrastructure around the model
Keep the model.
Lose the stack overhead.
before the model file
package tree, or container required
OpenAI- or Anthropic-style clients
Runtime Lite reduces the software footprint around local inference. The operational benefit depends on your model, hardware, concurrency, and present stack, so a production evaluation should measure capacity and throughput on the workload you intend to run.
Start with the inference engine
Local inference.
Without platform overhead.
Runtime Lite isolates the inference engine used within the wider Parta stack. It is intended for developers and teams that need a small local server first, without identity, audit, or policy services. Those controls are available only in the relevant commercial Runtime tiers.
No identity layer / No audit chain / No policy engine
Compatible by design
Speaks the formats
you already use.
OpenAI-compatible and Anthropic-compatible chat completion endpoints, so many existing clients and SDKs can connect to your local server with limited integration work, subject to the endpoints and features they use. Vision support is included for models that accept image inputs.
OpenAI-compatible / Anthropic-compatible / Vision included
Evaluate without changing the machine
Download, unzip,
run.
One statically linked executable and one configuration file. No admin rights, no installer, no background service to manage. Point the config at a GGUF model file already on your machine.
Bring your own model / No account / Local by design
Roadmap
Windows and Linux.
Vulkan next.
CUDA builds are available for both Windows and Linux. Vulkan is on the roadmap for hardware outside NVIDIA.
Windows CUDA / Linux CUDA / Vulkan (soon)
From evaluation to licensed production
Lite Free.
Lite.
Lite Free lets an individual or technical team validate model loading, API compatibility, vision inputs, and performance on existing hardware before considering a commercial deployment.
For production use, Runtime Lite provides the same focused inference role under a commercial licence. Organisations that also need workflow-level auditability or identity-backed provenance can evaluate Runtime or Runtime Pro.
Portable executable / No installer, Python, or Docker / Bring your own GGUF model
Free · Available now
Lite Free
Commercial · Contact us
Lite
For governed and regulated environments
Runtime.
Runtime Pro.
Production teams may need more than a local endpoint. Runtime is designed to verify origin, session and execution authority before protected inference resources are allocated. Governed deployments can receive resolved model context, authorised knowledge, execution-scoped adapters and controlled seeds from the wider Parta platform.
Runtime records defined execution events, outputs and provenance for later review, while Runtime Pro extends that model with identity-backed verification across nodes and services. These controls support governance and reconstruction; they do not replace risk assessment, professional review or compliance obligations.
Pre-inference verification / Controlled adapters and context / Execution records / Patent-pending architecture
Commercial · Contact us
Runtime
Commercial · Contact us
Runtime Pro
Download Lite Free and verify.
CUDA · v0.1.0 · Personal, educational and evaluation use
Requires an NVIDIA GPU with a CUDA 13-compatible driver. Vulkan build for other hardware in progress.
Download the archive and check its SHA256 checksum against the value below before running it.
Questions
Choose the right starting point
What is Parta Runtime Lite?
A single, portable executable that runs a local AI inference server on your own machine. It is a stripped-down version of Parta Runtime, the inference engine behind Parta Assist, with the identity layer, audit chain and policy engine removed.
Which API formats does it support?
OpenAI-compatible and Anthropic-compatible chat completion formats are available, so many existing clients and SDKs can connect with limited integration work. Compatibility depends on the endpoints and features each client uses. Vision is available for supported models.
Does it include identity, audit, or policy features?
No. Runtime Lite is intentionally stripped of everything that makes up production Parta Runtime. It is a plain inference server only.
What hardware does it require?
CUDA builds are available for Windows and Linux, requiring an NVIDIA GPU with a working driver. Vulkan is on the roadmap for other hardware.
How do I run it?
Download the zip, extract it anywhere on disk, point the included configuration file at a GGUF model file you already have, and run the executable. No installer, no admin rights, no background service.
Does it send any data anywhere?
Runtime Lite Free is designed to process prompts and responses locally and not transmit them to Parta. Your own network configuration, clients, model sources, and surrounding systems remain part of the deployment boundary you control.
What happens before governed inference begins?
In governed Runtime deployments, origin, session, scope and execution authority can be checked before protected inference resources are allocated. The wider Parta platform can then supply the authorised execution context, adapters, knowledge references and seed controls required for that task.
What is the difference between the four tiers?
Runtime Lite Free is for personal, educational, and evaluation use. Runtime Lite provides the inference engine and endpoints under a commercial production licence. Runtime adds signed, tamper-evident execution records. Runtime Pro adds identity-backed provenance. Contact us to evaluate any commercial tier.
Overview
Build the business case for a smaller inference stack.
Get the English infrastructure brief covering footprint, deployment choices, tier boundaries, and the questions to test against your own workload. We use your details to provide access and respond to relevant commercial interest.
Commercial evaluation and technical contact
Evaluate Runtime on your workload.
Tell us what you need to run, control, or verify
For a commercial evaluation, tell us your intended workload, hardware, integration format, adapter and knowledge requirements, and what must be verified before and after inference. The same form also reaches technical support for installation issues, model compatibility, bugs and performance questions.