Choosing an AI model for Familiar.

Which model should you pick?

The short version: connect a current flagship from Anthropic, OpenAI, or Google and you will not have to think about this again. A capable mid-tier model covers a tighter budget and carries most of a normal session. A strong local model runs offline if you want your world to stay on your machine. You bring your own AI, so all three are yours to take, and you can change your mind between sessions.

Familiar names the models it can reach in the picker itself, which is the honest place for it. Line-ups change every few months, so a list printed on a marketing page is stale before you read it. What does not change is the shape of the choice.

The model sets the ceiling. On D&D 5e (2024), Familiar's code holds the damage, the conditions and the dice steady, so a weak model cannot break the maths. What it can do is bore you. The words are the model's own: whether the read-aloud lands, whether your innkeeper still sounds like himself in hour three, whether the recap is worth reading. That, and the judgement when two conditions collide, is what a stronger model buys.

• MCP vs API: which to chooseWho you pay, and what a session actually costs on each route.
• Running a local AI modelThe Local AI tab, the four runners, and what local play asks of your PC.

Where do you actually get one?

Three routes, and they hand you different amounts of choice. Worth knowing before you sign up for anything, because one of them quietly costs you a setting.

Whichever you take, none of it is a commitment. Your campaign lives in Foundry journals and Familiar's memory, not in the model, so swapping provider halfway through a campaign costs you nothing but the time to paste a key.

• One key for everything: OpenRouter reaches hundreds of models, and you swap between them from a dropdown without signing up anywhere else. The shortest path from nothing to playing.
• A direct provider key: reaches that provider's own catalogue, and on Anthropic it also unlocks the full Thinking ladder. OpenRouter only reports whether a model can reason, not how finely, so routing Claude through it gives you Low, Medium and High and nothing above.
• Over MCP: you do not pick a model in Familiar at all. Your client picks, and Familiar runs the tools it calls.

• Connect Claude, ChatGPT, or Gemini over MCPPer-client setup, and the plan each one needs.
• How to add AI to Foundry VTTThe install and API-key path, start to finish.

Where the model actually shows up

Think of the model as the brain that executes your prep. The journals, statblocks, and encounters are already on the page; the model is what reads them and works the tools to make them happen at the table. Prep fixes what runs. The model sets how well it runs.

One turn of combat is a multi-step tool sequence: roll initiative, set the order, apply a condition, force a save, update the canvas, then play the monster turn. Every step is a chance to slip, and the slips compound. A model that gets each step right nineteen times in twenty still comes through that six-step chain clean only about three nights in four.

On D&D 5e (2024) Familiar takes some of that risk off the model: the damage, the conditions and the dice run in code, so the arithmetic holds up whatever you connect. What the model still owns in that chain is the order and the reading: which tool to reach for next, whether a rule applies at all, whether a ruling survives two stacked conditions. That is why the model shows up most in combat.

On other Foundry systems your own system resolves all of that, so more of the night rides on the model itself: the language, the lookups, and the calls you hand back to the table.

Speed is the other half of the trade, and pricing pages never mention it. A flagship thinking hard is dead air at a table with people sitting in it. Playing solo you will barely notice. With four players watching you wait, you will.

• Combat & AI

• What Familiar does on your systemWhich guardrails run where, drawn per system.

Give it room to think

Reasoning effort, what Familiar calls Thinking, is how hard the model deliberates before it acts. Turn it down and the AI answers off the top of its head. Turn it up and it works the problem through first: reads the rule, checks the board, plans the sequence, then moves. You set the dial, and the model spends the extra effort only where you point it.

Familiar ships this on High. That is already the right place for combat and hard rules, so if you never touch the setting, you are where the fiddly nights want you to be.

Which rungs you see depends on the model you connected. Claude Opus offers the whole ladder, Off through Max. Most others give you Low, Medium and High and stop there, so on Gemini Pro, Grok, Mistral or DeepSeek there is no way to switch reasoning off at all. Familiar only shows the levels your model genuinely accepts, so a shorter dropdown is the model's limit rather than a missing feature.

1. 1

   Combat & tricky rules: High

   Where Familiar starts out, and where it should stay for the multi-step bookkeeping. This is the work that earns the extra thinking.

2. 2

   Roleplay & exploration: Medium

   One rung down from the shipped default. NPC banter, reading a journal, narrating a room. Enough deliberation to stay in character and on the page, without spending it on a conversation.

3. 3

   Quick lookups & recaps: Low, or Off where you have it

   Fetch a stat block, restate a rule, summarise last session. Off is the fastest and cheapest, but only Claude and GPT offer it; everywhere else Low is the floor.

Thinking is a dial you control, not a promise the model keeps. Turning it up improves the odds on a hard sequence; it does not guarantee the answer. Past a point the model over-thinks a simple ask and you pay for tokens you did not need. Give it enough room for the task in front of it, and no more.

• What a session costsMore thinking means more tokens. Here is what that adds up to.

Is it the model, the dial, or your prep?

Before you go and buy a bigger model, work out what actually went wrong. Most of what looks like a weak model is not one, and a flagship will not fix a page you never wrote.

• Staying consistent across sessionsHow the AI reads from your notes instead of a fading window.
• Prepping a published adventure for AIWhat goes in the journals, the sheets, and the one-page outline.

A starting point

If you would rather not tune any of this: connect a current flagship, leave Thinking where Familiar puts it, and play a session. That is close to the out-of-the-box setup already, and it suits most tables.

Adjust from there. Drop the dial when a night is all talk, and only reach for a cheaper model once the bill starts to bother you.

• Install Familiar in Foundry
• MCP vs API: which is cheaper for your table
• Staying consistent across sessions
• Not sure what to connect? Ask in Discord