Smith · planned AI features

An assistant for your modded setup.

We’re building Smith into Modsmith to help with crashes, conflicts, translations and mod choices. Local checks are part of the free manager. Cloud AI is still in development and is not available to use or purchase.

Plans and allowances Where Smith stops

Smithcrash-2026-09-12-21-40-05.log

Crash Logger SSEEXCEPTION_ACCESS_VIOLATIONfaulted in clothphys_ext.dll+0x1A2F2 modules · 214 plugins

The crash is inside Cloth Physics Extended. That DLL was built for game runtime 1.5.97, and you are running 1.6.1170. The mod page lists a build for your version.

ProposalBacked up · reversible

Update Cloth Physics Extended to 2.3, built for 1.6.1170

Opens the file on Nexus for you to download; it installs as a new version when it lands. The current version stays in the library.

Smith proposes, Modsmith executes.

Interactive preview of the planned workflow. The example is fictional; these buttons only change this illustration.

Planned capabilities

Five ways Smith can help.

These describe the intended workflow, not features available today. Tasks that change files must end in a plan you can review before applying it. Availability will be stated for each release.

Crash diagnosis

Start with a supported crash log. Local patterns suggest checks without an AI request. Cloud diagnosis is intended to explain the parsed log, identify likely causes and suggest actions such as checking a version or a missing dependency. A log alone may not establish the cause.

planned · diagnosis allowance

Conflict patches

When mods edit the same supported records, Smith is intended to propose how to combine the changes. Modsmith will validate a supported proposal before building a patch. Some conflicts require a manual choice or cannot be combined safely.

planned · patch allowance

Translation

Choose supported mod text and a language. The planned translation workflow preserves identifiers and formatting and keeps the original files available. Review names, terminology and meaning before using or sharing a translation.

planned · measured in words

Mod picks

Describe what you want from a setup. Smith is intended to suggest mods, explain the choices and check known dependencies against your build. Compatibility still needs testing; you choose what to download through your own Nexus account.

planned · selection allowance

Build with Smith

Work from a new idea or an existing setup: discuss additions, dependencies, conflicts and load order. Supported changes go through a reviewable plan and the manager’s own commands. This workflow is planned for a later stage.

planned · Pro limited · Max full

Local checks

Local checks, without a subscription.

The development build already has local checks for file overlaps, plugin dependencies and deployment drift. Supported crash logs can be matched against known patterns. These checks run on your machine and do not require a cloud model.

Who overrides whom. The mods with the most overlapping files, with wins and losses.
What the game will load. The plugin order with engine indices and the mod each plugin comes from.
What was checked. Mod folders readable, file overlaps, missing masters, fully overridden mods.
Build drift. A warning when what is deployed is not what is selected.
Crash signatures. Recognised patterns with suggested checks, computed offline.

The Smith section: local checks listing build state, library and load order slots, a top overlap card with the quoted mod names, who overrides whom, and what was checked. — Development preview of local checks in the Smith section.

Planned model selection

Choose a role or a specific model.

The planned catalog includes models from Anthropic, Google and OpenAI. Model choice will not be restricted by subscription tier; heavier models will use allowance faster. The available catalog will be published when the service launches.

Auto

Select a model suited to the task.

Spark

Prioritise speed for short questions and quick checks.

Anvil

Balance speed and depth for everyday tasks.

Forge

Request more involved analysis. It can take longer and use more allowance.

Roles describe a preference for speed or depth. Their model mapping may change as the catalog evolves. Model availability and request limits will be shown in the app.

Where Smith stops

Smith proposes. Modsmith executes.

These boundaries guide the implementation. They keep model output separate from the code that changes files.

Proposes, never executes: A model's answer is a structured proposal. Deterministic code validates it and runs an existing command, the same one you would click. The model never writes a plugin file and never touches the game folder.
Local first: Build checks, overlaps, missing masters, slot counts and crash signatures run on your machine, free, for everyone, even when Smith's model is off.
AI requests need your action: The intended flow shows what the feature will send before you first enable processing. Different tasks need different data: a crash diagnosis uses parsed fields; translation needs the text. Review names and text for personal information.
Untrusted text is data: Descriptions and logs are treated as input to analyse, not as authority to run commands. Model proposals still need validation and your approval. These boundaries reduce risk; they do not guarantee a correct answer.
Your key stays home: The Nexus API key is never part of any request to Smith and never stored on our side.
Confirmations written by code: When a proposal changes something, the confirmation text is built by the engine from the real plan: what changes, what is backed up, what can be undone.

What is sent

Shown first, sent second.

The crash-request preview uses parsed fields: game version, exception, modules, stack frames and plugin names. It omits the raw log and is designed to remove recognised personal paths. Names and text may still reveal something about your setup, so review the request.

Translation and build planning need different inputs. Before those services launch, the privacy policy and request screen will identify the data and providers involved. Viewing this illustration sends no AI request.

Crash diagnosisIllustration

Review before sending

The planned request review shows the context for this task.

Parsed crash details: Game version, exception and stack frames
Build context: Relevant module and plugin names
Check for private information: Names and text can reveal details about your setup. Review the actual request before allowing it.

No request is sent by this preview.

Illustration of the planned request review, not a live AI request.

Local checks are free. Cloud plans come later.

See the planned allowances and how they fit different modding workflows. Subscriptions are not on sale.

See the plans Release status