Msty Nexus 0.2.0: A Clearer Gateway for Local Runtimes, Routes, and Clients

Msty Nexus 0.2.0 makes the Nexus gateway easier to route through, inspect, and operate.

This release adds Smart Route for controlled request routing, local runtime discovery for reusing existing services, preset-scoped acceleration for explicit speedups, token-level usage reporting, clearer endpoints, renameable client tokens, smoother onboarding, and cleaner model labeling.

Smart Route Gives Nexus A Routing Layer

Smart Route is the main addition in Msty Nexus 0.2.0.

It lets Nexus expose a route pack as a stable @route/{slug} model identifier, then resolve each generation request to a concrete target model through the gateway. Connected clients can call one consistent value while Nexus keeps the selected target and fallback path inspectable.

Smart Routes help with:

• Reducing repeated manual model selection.
• Making routing choices easier to review.
• Keeping fallback behavior explicit.
• Giving clients one consistent model value to call.
• Making multi-model workflows easier to organize.

Existing Local Runtimes Are Easier To Reuse

Msty Nexus 0.2.0 detects supported local runtimes that are already running so that they can be easily added to Nexus.

This includes runtimes such as Ollama, llama.cpp, LM Studio, vLLM, and LiteLLM. Discovery is read-only, stays on the local machine, and returns safe details such as detected candidates, already registered endpoints, model counts, and model previews.

When a detected runtime is registered, Nexus verifies the endpoint, adds it as a local provider, enables automatic model listing, and does not store credentials for that runtime.

This makes existing local infrastructure easier to reuse within Nexus.

Speculative Decoding

Speculative decoding is now available and can be added as presets. Speculative decoding can drastically speed up text generation of LLMs.

When a request calls the target model directly, Nexus leaves it unaccelerated. Acceleration only applies when the request uses an accelerated preset slug, where Nexus can validate the profile and route through the preset-specific runtime configuration.

For llama.cpp presets, acceleration can pair the chat model with a smaller compatible helper model. For Nexus-managed MLX presets, Nexus supports compatible helper-model acceleration and native MTP acceleration where available.

For speed-focused workflows, the preset carries the acceleration intent, compatibility can be checked before use, the profile can be applied or disabled, and usage stats can show whether the faster path is reducing response time.

Faster responses are easier to evaluate when the target model, helper model, and acceleration profile remain explicit in the preset.

Usage Reporting Adds Client Attribution

Nexus 0.2.0 gives usage reporting more detail across connected clients, routed requests, and accelerated presets.

With token-level reporting, each gateway request can be tied to the local Nexus token that authenticated it. Usage can be filtered or grouped by tokenId, which gives Nexus a stable way to connect activity back to the client that made the request.

Route and acceleration reporting add more operational detail. Nexus can show route-pack activity, selected targets, fallback reasons, provider and model usage, accelerated preset activity, token volume, and tool-call counts.

Client Tokens Are Easier To Review

Client token labels can now be renamed.

Renameable labels keep token lists easier to review and make token-level usage reporting more useful, because each label can reflect the client, integration, or workflow it represents now.

Model Sources Are Easier To Read

Model and provider labels help to make models easier to identify when reviewing model lists.

This release improves how Nexus names local runtime providers, detected local APIs, hosted providers, runtime catalog entries, Hugging Face sources, and imported files making it easier to understand where a model comes from and how Nexus will route to it.

Cleaner labeling reduces confusion during everyday model selection, especially when several sources are connected.

Setup Paths And Endpoints Are Clearer

Setup now separates local runtime work from hosted provider configuration, guiding users toward installing or reusing a runtime, finding one already running, or adding a provider.

Connected-client onboarding is clearer as well, including workflows for tools such as Vibe CLI and Msty Claw, where users need the right endpoint and token path before requests can flow through Nexus.

Endpoint handling also improves, with the gateway panel distinguishing Gateway URL from OpenAI-compatible URL, supporting direct copy actions, and pointing users toward client token management when authentication is needed.

Together, these changes reduce friction in places where precision matters most, including runtime selection, provider configuration, endpoint choice, and client authentication.

Nexus 0.2.0 Smooths Everyday Workflows

In addition to the above, we also released the following enhancements and fixes:

• Smart Route pack availability validation is stricter.
• Provider models stay out of public selectors when they should not be shown.
• Non-GGUF results are filtered out for llama.cpp installs.
• Duplicate preset behavior is cleaned up.
• Runtime restarts are more reliable after network exposure changes.
• Model search cards and local model workflows are easier to use.
• App focus, popup placement, tray icons, and packaging behavior are more polished.

A Clearer Operating Layer For Nexus

The release centers on Smart Route, which adds a structured way to send generation requests through route packs with explicit fallback behavior. Runtime discovery, preset-scoped acceleration, token-level usage reporting, clearer endpoints, renameable client tokens, and cleaner model labels round out the workflow around it.

Update to Msty Nexus 0.2.0 to try Smart Route, scan for runtimes already running on your machine, review the client tokens and model paths connected through Nexus, and see the full 0.2.0 changelog for the complete list of changes.