Multi-LLM

Twelve ways to run the repair model

Your key, your machine, your CLI agent, or the engine built into WeftKit. Whatever you pick, the model's role is the same and strictly bounded: select one safe, locally validated recovery candidate — nothing more.

How requests are priced

API providers

BYOK API

Bring your own key. WeftKit talks to each provider directly with the API key you supply — there is no WeftKit proxy or middleman account.

Google Gemini

Your API key

Anthropic Claude

Your API key

OpenAI

Your API key

DeepSeek

Your API key

Alibaba Qwen

Your API key

Kimi / Moonshot

Your API key

Data boundary: requests go straight to the provider under your key. WeftKit sends only the bounded repair packet, requests stateless, no-storage handling where the provider's API supports it (for example Gemini's storage-disabled setting), and never includes your secrets. Retention is otherwise governed by your agreement with the provider.

Local runtimes

Local runtime

Run the repair model on your own hardware. Point WeftKit at a loopback endpoint and pick any model the runtime serves.

Ollama

No key — local endpoint

LM Studio

No key — local endpoint

Data boundary: with the default loopback endpoint, nothing leaves your machine. Repair requests go to a local endpoint you run; no key is required.

CLI agents

CLI agent

Already pay for a coding agent? WeftKit can hand the bounded repair prompt to a CLI agent installed on your machine and validate what comes back.

Claude Code

The agent’s own login

Gemini CLI

The agent’s own login

Codex

The agent’s own login

Data boundary: the agent runs as a locked-down subprocess with a cleared environment and uses its own authentication — WeftKit never sees or stores those credentials. Output is schema-validated before anything is accepted.

Embedded engine

Built in

No account, no key, no separate runtime. WeftKit ships an embedded llama.cpp engine that runs Gemma 3 4B on your hardware — a one-time model download, verified against pinned checksums.

Embedded (llama.cpp)

No key — Gemma 3 4B, fully local

Data boundary: the engine runs inside WeftKit as a llama.cpp server on loopback. Nothing leaves your machine — repair requests never touch the network.

Priced, always

Every model has a price card — even yours

Model choice never changes the accounting. Whatever runs the repair, the request is priced and recorded the same way.

Bundled default prices for well-known models, with per-provider source URLs

Unknown models — including local and CLI ones — fall back to $3 per million input tokens and $15 per million output tokens, so nothing is ever “free” by omission

You can override the price of any model, or register your own

Hard per-run and monthly budgets apply to every provider kind

Read the cost-transparency page

Pick a model. Keep control.

Bring your own key, run a local model, or reuse a CLI agent you already pay for — the boundaries stay the same.

Get the app See how it works