What MatterWave is

MatterWave is a hosted chat product and the platform under it: language models served behind one OpenAI-compatible API, a set of clients that all share the same conversations, and an organisation model with seats, budgets, API keys and single sign-on.

Chat is the first app on the platform, not the whole of it. The same /api routes the web app uses are the ones the terminal client, the VS Code extension and the desktop app call, and the same /v1 surface the chat runs on is what the SDKs wrap.

The models

MatterWave serves the GRT family. The three names are tiers of the same family with different list prices, and each one spends your plan's allowance at a different rate:

Model	Input, per 1M tokens	Output, per 1M tokens	Allowance multiplier
GRT Nano (grt-nano)	$0.10	$0.40	1×
GRT Nano Pro (grt-nano-pro)	$0.50	$2.00	5×
GRT Nano Max (grt-nano-max)	$2.00	$8.00	20×

Every plan includes the deployment's house model. Paid plans add the dearer tiers; see Plans and pricing. Partner models from other providers may be offered on particular plans; they are opt-in per plan and never appear on a plan that did not name them.

The clients

• Web app at app.matterwave.ai: conversations, branches, attachments, model switching, custom instructions, usage and budget, organisation and member management, API keys.
• Terminal (matterwave-tui): the same conversations in a terminal, with the model picker, branches and a status line. Docs · Download.
• VS Code: a chat view in the side bar that can send the editor's selection or a file as context. Docs · Download.
• Desktop: the web app in a native window with a sign-in that survives restarts and matterwave:// links. Docs · Download.
• Raspberry Pi: the API, the database and a CPU decoder for grt-nano on one small box, with no cloud in the loop. Docs · Image tags.

The API and SDKs

/v1 speaks the OpenAI chat-completions wire format: chat/completions with streaming, models, and embeddings where a model supports it. A key is created once in Settings and scoped (chat:read, chat:write). The Python and TypeScript SDKs add typed responses, a streaming iterator, retries that honour Retry-After, and the server's request id on every error.

Within v1 changes are additive: no field is removed or retyped, and a deprecation carries Deprecation and Sunset headers with twelve months of notice. The API changelog lists what a client written against /v1 can see.

Organisations

One email is one account in one organisation. An organisation's administrators invite members, attach a single sign-on provider on the Team plan, set spending limits, issue and revoke API keys and read usage. Plan changes never delete anything: a downgrade that leaves an organisation over its new plan's limits makes it read-only over the limit, and upgrading again restores writes on the next request.

What it is not

It runs no tools and edits no files on your machine: the clients are chat clients, not coding agents. There is no send queue in any client: a message is sent while you watch or not at all.