Every model you pay for, in one workspace you own
Max Harness runs on your own server. Connect your API keys for OpenAI, Claude, Gemini, OpenRouter and 21 other providers, then chat, compare answers and run tool-using agents from a single browser tab.
No bundled model access and no markup. You pay each provider directly for what you use.
The timeline leaves no slack between the data migration and the freeze date.
Two of the five workstreams share one owner, which is the likeliest point of delay.
A complete chat workspace, without handing your history to another service
The things you expect from a desktop AI app are here. The difference is where it runs and who holds the keys.
Switch models mid-conversation
Change provider or model at any point. Every reply is labelled with the model that wrote it, so a long thread stays readable.
Compare answers side by side
Send one prompt to up to four models, read the replies next to each other, then continue with the one you prefer.
Projects and custom assistants
Group chats into projects with shared instructions. Save assistants with their own prompt, preferred model and settings.
Files and images
Attach PDFs, Word documents, code and images. Max Harness tells you up front when the selected model cannot read a file type.
Agent tools with approval
Let a model search the web, read pages, save notes to memory or create files. You decide which tools run freely and which ask first.
Usage you can check
Token counts come from the provider. Cost appears only when a reliable price exists, and it is always marked as an estimate.
25 provider presets, plus any compatible API
Pick a provider, paste your key, and the endpoint and model list are set up for you. Each preset was checked against the provider's own documentation.
Your keys stay on your server
Requests go from your server straight to the provider you chose. Nothing passes through a third party and no key is ever sent to a browser.
- Encrypted credentials. API keys are encrypted at rest with your installation's own key and shown again only as their last four characters.
- Private by account. Conversations, files and provider connections belong to one user. Administrators manage the site, not other people's chats.
- Export and delete. Download everything as a ZIP, remove a key, clear your history or delete the account, all from the settings screen.
- Optional memory. Off until you turn it on. Every saved item is visible, editable and removable.
- Two-factor sign-in. Authenticator app codes with recovery codes, plus a list of active sessions you can revoke.
From ZIP file to first reply in three steps
Upload and install
Upload the release ZIP to a PHP 8.4 site, open it in a browser and enter your database details. No terminal, Composer or npm required.
Connect a provider
Choose a preset, paste the API key from that provider's console and press Test. The model list loads automatically where the provider offers one.
Start working
Open a chat, pick a model and send a message. Add projects, assistants and tools when you need them.
Practical notes on running your own AI workspace
Why agents in Max Harness ask before they act
Tool-using models are useful and also easy to mislead. This is the reasoning behind the short tool list, the three permission levels and the approval step.
How cost estimates work, and why some replies show none
Max Harness shows a dollar figure only when it can stand behind the price. This is where the numbers come from and where they can differ from your invoice.
API keys and chat subscriptions are not the same thing
A ChatGPT or Claude subscription does not unlock the API. Here is how the two differ, what an API key costs in practice and how to keep spending under control.
Bring your keys. Keep your conversations.
Set up a workspace on your own domain and use the models you already pay for from one place.