> ## Documentation Index > Fetch the complete documentation index at: https://linkly.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Official Model Service > Linkly AI's own cloud model service: sign in and start chatting, with no API key of your own to supply and a single shared quota pool. ## What the Model Service Is The [Chatbot](/docs/en/use-chatbot) in the desktop app needs a language model to drive it. Traditionally that meant registering with some provider, enabling billing, creating an API key, and coming back to fill in a pile of configuration. **The model service** is Linkly AI's own cloud model access: **sign in to your account and start chatting, with no API key of your own to supply and no card to link**. It's a curated set of the models that performed best in our testing. Sign in and go — free accounts get quota too, and no credit card is involved Not per token, not per request — every model draws on the same pool The model service covers **chat inference** and nothing else. Document indexing, text recognition (OCR), and speech transcription (ASR) **all run on your own machine** and have nothing to do with this service — retrieval keeps working perfectly even if you never touch the official models. *** ## Getting Started Open **Settings → Account** and click **Sign In**. Your browser opens the authorization page, and once it's done you're returned to the app automatically — the API key is fetched and stored by the app, with no copy-pasting on your part. If the browser fails to open, expand **Optional: connect your account by setting an API key manually** at the bottom of the account page and paste in the `lkai_` key from the [dashboard](https://linkly.ai/dashboard/integrations); the result is identical. Open **Settings → AI → Providers**; the **Linkly Official** card should read **Ready**. If it reads **Needs setup**, you aren't signed in yet — go back to the previous step. Open the card's settings dialog: **Refresh** (once models are listed) or **Add models** (when none are) pulls the currently available model list from the server, so a single click syncs any catalog updates across. Open the Chatbot. The model picker sits in **the toolbar to the right of the composer**, grouped by provider. Official models carry a multiplier badge; hover it to see roughly how much quota each request consumes. **Settings → AI** is also where you set the model used by default for new chats, along with a pinned response language. *** ## Which Models Are Available The official catalog comes in two tiers: * **Basic models** — available to everyone, with free quota included; good for everyday questions and for trying things out. * **Frontier models** — require a Pro subscription, and are more capable. **Which models exist, and their individual multipliers and context windows, are whatever the in-app model picker actually shows.** The catalog is delivered by the server and shifts with provider pricing and new model releases. Free accounts **don't see the frontier models in the list at all** (the server filters them out), so you never pick one only to find you can't use it. *** ## Quota This section is worth two minutes of your time — it decides how much you actually get to use. ### One Shared Pool, Not Per Token and Not Per Request The official models are **not billed per token, and no model gets its own separate allowance**. Every model draws on **the same pool**, each model has its own **multiplier**, and one request deducts the corresponding amount: > **1 unit of quota = one request to a model with a ×1 multiplier.** The lower the multiplier, the less quota a model burns: one request to a ×0.5 model costs half a unit, one request to a ×4 model costs four. Each model's multiplier is printed on its badge in the model picker, and hovering it shows roughly how much quota it consumes. **Use the basic models for everyday questions and switch to a frontier model only when you genuinely need deep reasoning** — that's the most economical pattern. Out of the same pool, a several-fold difference in multiplier means a several-fold difference in how many turns you get. ### Three Parallel Time Windows The pool has **three time windows: 5 hours, 7 days, and 30 days**, each counted independently, and **filling any one of them gets you rate-limited** — even when the other two still have room. The 5-hour window stops you burning everything in one sitting; the 7-day and 30-day windows constrain overall usage. A Pro account's allowance is substantially higher than a free account's in all three windows. **The exact allowance per tier is whatever the in-app quota panel and the [website](https://linkly.ai/pricing) say.** Click the **Model quota** button at the bottom of the Chatbot sidebar to see live headroom in all three windows and each one's reset time. The windows are **fixed windows**, much like ChatGPT's and Claude's: the clock starts on your first billable request in a given cycle, and when it's up **the whole allowance is restored at once**. ### What Costs Quota and What Doesn't | Action | Costs quota | | ------------------------------------------------------------------ | ----------- | | You send a new message (starting a new turn) | **Yes** | | The AI calls retrieval tools and reads documents during that turn | No | | Automatic chat title generation, and `/compact` context compaction | Yes | | Opening the quota panel | No | | Using a third-party or local model | No | **Tool calls not being billed separately** is an important design decision: in a single turn the AI might search seven or eight times and read a dozen files, and all of it shares the one deduction from the message you sent. So "how many questions did I ask today" is roughly "how much quota did I use", and you're never charged extra for the AI being thorough. *** ## Where to Check It, and What Happens When It Runs Out ### Where to look Two places: 1. The **Model quota** button (the gauge icon) at the bottom of the Chatbot sidebar. It opens the **Model quota** panel, with one row per window — **5h quota**, **7d quota**, **30d quota** — each showing **remaining / total** and **when it resets**. 2. The home page of the web [dashboard](https://linkly.ai/dashboard). When you're not signed in it reads `Sign in to your Linkly account to view your quota`; when you're offline it reads `Couldn't load quota — check your connection`. **The official models require an internet connection** — that's the most practical difference between them and a local model. ### What happens when it runs out Once a window is full you can **still send messages**, but that turn fails immediately and a notice appears in the conversation telling you exactly which window is full and when it recovers: | Window filled | What the notice says | | ------------- | ----------------------------------------------------------------------------- | | 5 hours | Linkly AI quota limit reached (**N** uses over **5**h). Resumes at **HH:MM**. | | 7 days | Weekly quota limit reached (**N** uses over 7 days). Resumes **date**. | | 30 days | Monthly quota limit reached (**N** uses over 30 days). Resumes **date**. | The bold parts are filled in by the app according to your actual tier: **N** is that window's total allowance, and what follows is the exact moment it recovers. So you never have to work out when you can carry on — the notice tells you outright. A few things worth knowing: * **Nothing you already have is lost.** The rate limit affects only that one request; past messages, the conversation list, and the search index are untouched. * **It's the whole pool that's limited, not one model.** All three windows draw on the same pool, so once you're limited, **switching to a lower-multiplier official model doesn't help** — a low multiplier changes *how much each request costs*, not *whether you can send one*. * **Third-party and local models are entirely unaffected.** They take a different route and never touch the official quota (see the next section). * **There's nothing to request** — the full allowance is restored automatically when the window is up. So when you're rate-limited there are really only two routes: **wait for the window to reset**, or **switch to a third-party or local model** and carry on. The way to actually save quota is to build the habit before you hit the limit — basic models for everyday questions, frontier models only when you need deep reasoning. *** ## You Don't Have to Use the Official Service (BYOK) The official model service is optional. The Chatbot supports **any OpenAI-compatible endpoint**, added under **Settings → AI → Providers → Custom provider**. 12 presets are offered when adding one, to fill in the address quickly: | Group | Presets | | ----- | ------------------------------------------- | | Cloud | OpenAI, DeepSeek, Mistral, Groq, OpenRouter | | China | Qwen, MiniMax, Moonshot, Zhipu, SiliconFlow | | Local | Ollama, LM Studio | Presets **only prefill the Base URL**; they don't include a model list. Once the address is in, you can auto-discover models from that endpoint's `/v1/models`, or enter model ids by hand. Local servers (Ollama / LM Studio) usually **need no API key** — just leave it blank. ### How Do You Choose a Model Source? | Approach | Bring your own key | Quota / billing | Where data goes | Who it suits | | ------------------------ | ------------------ | ----------------------------------------- | ----------------------------------------------------- | ------------------------------------------------------- | | **Linkly Official** | No | Shared quota pool, free tier included | Forwarded to the provider via Linkly's servers | You want to start now and not fiddle with configuration | | **Third-party endpoint** | Yes | You settle up with that provider directly | Straight to that provider's API, never through Linkly | You already have API credit, or want a specific model | | **Local model** | No | Free, at the cost of your own compute | **Never leaves your machine** | Zero tolerance for data leaving the device | If a custom provider doesn't return a `context_length` field (very common with local Ollama / LM Studio), the app has no idea how large the model's context window is — **the usage percentage can't be shown and automatic compaction won't fire**. With those models, keep an eye on the conversation length yourself and run `/compact` or start a new chat when the time comes. See [Using the Chatbot](/docs/en/use-chatbot) for details. *** ## Where Data Goes, and Privacy Being straightforward about where data ends up on each path: | Capability | Where data goes | | --------------------------------------- | -------------------------------------------------------------------------------------- | | **Language model · Linkly AI Official** | Data first reaches Linkly AI's servers; we then forward it to a curated model provider | | **Language model · Direct third-party** | Straight to that provider's API, never through Linkly | | **Language model · Local** | Switch to Ollama or a custom local model and conversations stay on your device | | **Embedding generation** | Runs locally; the model file is downloaded from the official CDN on first launch | | **OCR (text recognition)** | Runs locally; the model file is downloaded the first time OCR is used | Using the official model service means the content you send — **including the document excerpts the AI retrieved to answer you** — **passes through Linkly AI's servers before being forwarded to the upstream model provider**. That's the inherent cost of a relay service, and we won't pretend otherwise. The in-app **Settings → Data Privacy** page lists, live, which capabilities are currently sending data outward, so you can check for yourself at any time. Each chat request carries **the app name and version** in HTTP headers, used for provider attribution. **It contains no identifying information.** *** ## FAQ Yes, but you have to supply a model source yourself — add a third-party endpoint or a local Ollama / LM Studio under **Settings → AI**. The official model service is the only part that requires signing in; [search](/docs/en/use-launcher), indexing, MCP, and the other local capabilities need no account. No. The three windows (5 hours / 7 days / 30 days) are **all in effect at once and counted independently**, and filling any one of them rate-limits you. So heavy use in a short burst tends to hit the 5-hour window first, while slow steady use over time is more likely to run into the 30-day window. Read it straight off the multiplier — it's a direct ratio. A fivefold difference in multiplier means fivefold the number of turns out of the same pool. Every official model's badge in the model picker shows its multiplier, so a glance before switching is enough to estimate. No, these are **two entirely separate tracks**. Chat quota only counts model inference; [cloud libraries](/docs/en/use-cloud-library) and [remote access](/docs/en/use-in-remote) have their own rules and don't draw on it. Gemini's OpenAI-compatible endpoint requires the key to go in a `?key=` URL parameter rather than the standard `Authorization: Bearer` request header, and the current chat pipeline doesn't support that form of authentication. Gemini's tool-calling support also isn't quite there yet. We're watching closely and will add it once it is. For the official model service, signing in to the same account on the new device is all it takes — quota **follows the account** (all three windows are counted per account, not per device). The quota panel saying **Sign in to your Linkly account to view your quota** simply means this device isn't connected to an account yet; go to **Settings → Account** and sign in. If you're already signed in and still see it, the key has most likely expired — sign out and sign back in once. Custom provider addresses and keys are stored on the machine itself, so those do need adding again on a new computer. *** ## Further reading * [Using the Chatbot](/docs/en/use-chatbot) — the full workings of the chat window, `@` references, and context compaction * [Using Cloud Libraries](/docs/en/use-cloud-library) — take a library off your machine and keep it online around the clock * [Remote Access](/docs/en/use-in-remote) — let cloud AI reach back into the documents on your machine * [FAQ](/docs/en/faq) — other frequently asked questions * [Account](/docs/en/linkly-account) — sign-in methods, API keys, and Pro benefits