Set up Claude

MindMap Chat runs on your own hardware by default, through Ollama or LM Studio. Claude is optional. Choose it when you want a more capable model than your machine can run, and accept the trade: what the AI needs to answer leaves your device. This page is the setup; what that means for your data is on the privacy page, and the app asks you to agree before a single word is sent. OpenAI is the other option, and the two work the same way.

You need API credit, not a Claude subscription

This is the step that catches people out. API access is billed separately. A Claude Pro or Max plan does not include API credit — they are two different products with two different bills. Likewise, a token from claude setup-token is a Claude Code credential and cannot be used here; Anthropic permits third-party apps to authenticate with API keys only.

So the first thing to do is add a small amount of credit. You pay Anthropic per token, directly. MindMap Chat takes no commission, never sees your account, and shows you the token count of every session so the bill is not a surprise.

Create the key

Go to platform.claude.com and sign in. The same account as Claude.ai works.
Open Settings → Billing and add credit. A key with no credit behind it authenticates fine and then fails on the first message, which is a confusing way to find out.
Open Settings → API keys and press Create key.
Name it something you will recognise in a year — MindMap Chat — and choose an expiration. Never is reasonable for a key that stays on your own machine; a dated one means remembering to replace it.
Scope it to a single workspace if you are offered the choice. This is the step that saves you trouble later: a key tied to one workspace needs nothing else. A key that spans several has to name the workspace on every request — see below.
Copy the key. It starts sk-ant-api… and is shown once.

Put it in the app

Open Settings → AI and set Provider to Claude (Anthropic).
Read the dialog and accept it. Declining leaves your previous provider selected — nothing is sent, and nothing is changed.
Paste the key into API key.
Press Reload next to Model. This asks Anthropic which models your account can actually use, and flags any the app lists that it no longer offers.
Press Test connection. All four checks should go green: endpoint reachable, authentication accepted, model available, test reply received.

If it asks for a Workspace ID

A key that is not tied to one workspace has to say which workspace each request acts in, and the app will tell you so rather than failing with an HTTP error you cannot act on. Two ways out, and the second is tidier:

Open Settings → Workspaces in the Console, copy the value from the ID column — it starts wrkspc_ — and paste it into Workspace ID in the app.
Or create a new key scoped to a single workspace, and leave the field empty.

What it costs you, and how to keep it small

Two controls sit under Cost and size in the AI tab:

The session total, in tokens Anthropic itself reported — not an estimate. Your Console dashboard remains the authority on the bill.
Ask before sending more than, a ceiling on one request, 32k tokens by default. A large map is both a bill and the thing leaving your device; above this the request is refused and tells you the number, rather than being sent quietly. Set it to No limit if you would rather it did not.

There is also an optional Scrub identifiers before sending switch, off by default. It replaces email addresses, IP and MAC addresses, GUIDs, file paths and token-shaped strings with placeholders. It is defence in depth, not a guarantee — and because your map is what gets sent, the model then answers about the placeholder, so a map full of real addresses gets worse replies with it on.

Troubleshooting

What you see	What it means
Authentication failed (401), "API key is invalid"	The key reached Anthropic and was rejected. Check it was copied whole, that it is an API key from the Console rather than a `claude setup-token` credential, and that it has not expired or been deleted.
A workspace ID is required	Your key spans more than one workspace. Fill in Workspace ID, or create a key scoped to one workspace.
Access denied (403)	The key is valid but has no access to the model you picked. Try another model, or check the key's permissions in the Console.
Not found (404)	Anthropic does not have that model. Press Reload next to Model — the app will list what your account can actually use.
No API credit left	The account has run out. Add credit under Settings → Billing in the Console — nothing is wrong with the app or your key, and waiting will not clear it.
Rate limit reached (429) or overloaded (529)	Nothing is wrong with your setup. 429 is your account's rate limit and the message says how long to wait; 529 means Anthropic is busy. The app waits and retries a couple of times on its own before reporting it.

Who is responsible for what

MindMap Chat is not affiliated with Anthropic or OpenAI, and does not resell, host or intermediate their services. When you enable one, you are using your own account with that provider, and that is an agreement between you and them: their terms of service, privacy policy, acceptable-use policy, pricing and availability govern it. Any account, billing, quota or support matter is yours to take up with them directly — this app has no visibility into your account and cannot act on your behalf.

Warranty and liability are the licence's, not a second set of terms. This software ships under the Elastic License 2.0, whose No Liability section reads:

As far as the law allows, the software comes as is, without any warranty or condition, and the licensor will not be liable to you for any damages arising out of these terms or the use or nature of the software, under any kind of legal claim.

That covers anything that goes wrong when a provider is involved — outages, rate limits, models changed or withdrawn, refused or wrong answers, how the provider handles what is sent, a suspended account, or charges made to it — for the same reason it covers everything else.

Deleting a chat here deletes it here. Clear conversation removes the messages from this device's encrypted database and makes no network call — there is nothing to delete at the provider in the sense you might expect, because the API is stateless and keeps no conversation of its own. What it does keep is the requests already sent, for as long as its retention policy says. Nothing in this app can reach that; the provider's privacy policy governs it.

What the app does guarantee is narrow, and worth stating plainly: it will not send anything to a provider you have not selected and agreed to. The local runtimes remain the default and send nothing anywhere.

Read the provider's own policies before enabling this:

Anthropic — Consumer Terms, Commercial Terms, Privacy Policy, Usage Policy
OpenAI — Terms of use, Privacy policy, Usage policies

Changing your mind

Switch Provider back to Ollama or LM Studio at any time and nothing goes to Anthropic again. Your key stays in the app's encrypted database until you clear the field and save, and every map you have made stays exactly where it was — encrypted, on your machine.

Prefer to keep everything local? That path is fully supported and remains the default.