How to set up FarkiBot
FarkiBot is Farkitect’s built-in AI assistant. It answers questions about your models, writes documents from them, generates new metamodels and models, and rewrites models you already have.
FarkiBot works on a bring-your-own-key (BYOK) basis: you supply an API key for the AI provider of your choice, and Farkitect uses it on your behalf. There is nothing to buy from us, and no AI subscription bundled into Farkitect.
The privacy promise: your API key and your model content are sent directly from your browser to your chosen provider. Farkitect’s servers are never in the path — and you can verify that yourself in your browser’s network inspector. See How FarkiBot Works for why it’s built this way.
1. Open FarkiBot
Section titled “1. Open FarkiBot”Choose View > FarkiBot (Ctrl+Shift+K). The panel opens alongside Properties, Context, and Notes.
The first time you open it, FarkiBot has no key and shows a Set up FarkiBot… button. You can reach the same settings at any time from the gear icon in the FarkiBot toolbar.
2. Choose a provider
Section titled “2. Choose a provider”One key is the whole setup. Paste an OpenRouter key and you have every model — GPT, Claude, Gemini, Kimi, DeepSeek, Llama and 300+ more — behind a single credential. Signup is free, no card is required to start, and there are free models you can use at $0.
That’s the recommended path, and for most people it’s the only one they need.
Already pay another provider?
Section titled “Already pay another provider?”You do not need a second card. Add your existing Anthropic, OpenAI, or Google key inside OpenRouter > Settings > BYOK. Those models then bill to your existing account at your existing rate limits, while FarkiBot still holds only the one OpenRouter key.
The Advanced options
Section titled “The Advanced options”Under Advanced are direct provider connections, for two narrower cases:
| Advanced option | When you actually need it |
|---|---|
| Anthropic / Google Gemini | You want that provider reached directly, with OpenRouter out of the path entirely. This is the stronger privacy posture — the answer when a client or a security review asks who sees your model content. It is not needed merely to use Claude or Gemini. |
| Custom endpoint | Any OpenAI-compatible base URL — local models via Ollama or LM Studio, or your organisation’s internal LLM gateway. |
If you already have a direct provider configured, the Advanced section opens by itself so your key stays in plain sight.
There is no direct OpenAI option. OpenAI’s API does not permit browser connections, so it can’t be reached the way the others are. Add your OpenAI key inside OpenRouter > Settings > BYOK instead: GPT usage bills to your existing OpenAI account, and the same OpenRouter key still gets you every other model. Note that OpenAI API accounts hold no conversational history — that lives in ChatGPT, a separate product — so nothing is “left behind” by routing through OpenRouter.
3. Create a capped key
Section titled “3. Create a capped key”Whatever provider you choose, create a dedicated key with a spend limit for FarkiBot. Then the worst case of any mishap is a cap you set yourself.
- OpenRouter — create the key under Keys and set a credit limit (say $10, reset daily). Buying about $10 of credits once also unlocks 1,000 free-model requests per day. Skip auto-top-up: a fixed balance is a hard ceiling.
- Anthropic — create the key in a workspace (such as Default), not for the whole organisation, and set that workspace’s spend limit in the console. A key scoped to the organisation is refused: Anthropic expects a workspace to be named with every request made by one, and FarkiBot names none.
This is a five-minute step that makes every later decision reversible. Do it.
4. Paste and test
Section titled “4. Paste and test”Paste the key into FarkiBot Settings. It’s masked from the moment you paste it, and after saving only the last 4 characters ever display — there’s no reveal button, by design.
Click Test connection to confirm the key works before you rely on it. If the provider refuses, FarkiBot shows the provider’s own reason beside the error, such as a key that is scoped to the wrong thing or an account with no credit, so you can tell what to fix.
Once a key is saved, the LLM list in the FarkiBot panel loads that provider’s models. If you change provider, FarkiBot keeps the model you were using where the new provider offers it too (the same Claude model has a different name on OpenRouter and on Anthropic, and FarkiBot translates it). Where it does not, the list asks you to choose.
5. Decide how the key is stored
Section titled “5. Decide how the key is stored”| Option | Behaviour |
|---|---|
| Session only (default) | The key lives in memory and disappears when the tab closes. Paste it again next session. |
| Remember on this browser | The key is persisted, encrypted (AES-256) with a key your browser holds internally that no website code — including Farkitect’s — can export. |
It is never stored in the clear. Anyone inspecting this browser’s storage sees only ciphertext.
The honest caveat: no browser or desktop app can protect a secret from malicious code running as the app itself. Farkitect’s defences for that case are a strict Content-Security-Policy — a stolen key would have nowhere to be sent, because the browser refuses connections to any host not on the allow-list — and the capped key you created in step 3.
Where settings live
Section titled “Where settings live”FarkiBot settings are per browser, per device. They’re stored locally, not on your Farkitect account, so:
- Setting up FarkiBot on your laptop does not set it up on your desktop
- Your key is never synced to Farkitect’s servers, because it never reaches them
- Colleagues on the same project each bring their own key
Next steps
Section titled “Next steps”- How to ask FarkiBot about your model
- How to choose an LLM — which model for which job, and what it costs
- Working with FarkiBot — the full walkthrough