MaeVae and the AI vendors
MaeVae is a desktop console that runs your AI coding agents side by side and keeps each one in a pane and a profile of its own. Sign in once per profile, and nothing crosses between them: not the login, not the bill, not the memory. Each pane holds an agent’s own command-line tool, and each profile uses your own account with that agent’s vendor.
MaeVae earns from its plans for the app and, later, from the MaeVae compute plans. It never resells Claude, GPT or Gemini usage. Each AI vendor’s rules are respected.
Between you and each vendor
-
The account is yours
A profile is one identity: one agent, one account, one project folder. You sign in inside the pane, with the agent’s own login; some profiles use an API key you enter instead. MaeVae holds no agent account of its own.
-
Nothing added in between
MaeVae adds no account and no proxy between you and a vendor. The agent’s own tool talks to its vendor, or to the endpoint you set in its profile.
-
When MaeVae itself sends a request
For a Claude profile with the HUD on, the quota reader asks Anthropic’s usage endpoint, with that profile’s own login; the other agents’ figures are read from their own files on your computer. List models asks the endpoint you typed, only when you click it, or, at launch, when you paste an address into the profile’s Endpoint section. The Install and Update buttons run the agent’s own published install or update command on your computer. At launch, in the Store release, the Grok install button shows you the command to run instead of running it.
The Privacy page lists the app’s network activity.
-
Each vendor’s terms apply
Each vendor’s terms and privacy policy apply to what its agent sends. They are the vendor’s own, and this page does not restate them.
It drives each agent’s own command-line tool
MaeVae bundles none of the agents’ tools. You install the ones you use and put them on your PATH, and MaeVae starts each one in its own terminal pane.
- Claude Code
- Codex
- Gemini CLI
- Qwen Code
- aider
- Grok
- and a plain shell
What each agent reports, and what it needs
Seven rows: the six agents and a plain shell. Restricted mode is offered for three of them.
MaeVae drives every agent the same way, but the agents do not sign in the same way, report the same figures or tell MaeVae the same things. Here they are side by side.
The HUD column names the figures each agent reports, as the strip under its pane and the pane’s header show them; n/a marks a figure the agent does not report.
| Agent | Sign-in | What the HUD reports | Before the pane lights up | Restricted mode |
|---|---|---|---|---|
| Claude Code | Its own login inside the pane, after its one-time setup: a theme picker, a prompt to trust the folder, then a sign-in in your browser. Or an API key set in the profile. | model · cost · time · tokens · quota · contextThe most of any agent: lines changed too, and the 5-hour and weekly quota with their reset countdowns. Switched on per profile with the hud button on its rail entry. | No trust step. The same hud button turns on its attention queue. | Yes |
| Codex | Its own sign-in inside the pane, or an API key set in the profile. | model · tokens · time · context · quota cost n/aThe quota is the rate limits Codex reports, with their reset countdowns; where Codex reports a single 30-day limit, the strip shows it as Monthly. Codex does not report a cost. | Once, inside Codex: type /hooks and trust MaeVae’s two hooks. Codex asks again whenever the hook text changes, for example after a MaeVae update. Its glow and its approve and deny buttons belong to the profile, not to one pane. |
Yes |
| Gemini CLI | A Gemini API key from Google AI Studio, set in the profile. Google’s own sign-in inside the pane works only with a Workspace Code Assist license. | model · tokens · time · context cost n/a · quota n/aGemini CLI reports neither usage nor cost. | Once: trust the project folder Gemini CLI opened in. Its hooks are on by default and listed under /hooks panel. Close and reopen the pane after turning the glow on. |
Yes |
| Qwen Code | An API key, an Endpoint and a Model set in the profile: it needs all three to use the endpoint. Or its own /auth inside the pane. |
model · tokens · time · context count cost n/a · quota n/aThe context gauge only when Qwen Code knows the model’s window. It reports neither usage nor cost. | No trust step: close and reopen the pane, and the glow is live. | No |
| aider | No login: its API key, set in the profile, is its account. | model · tokens · time · cost context n/a · quota n/aThe last prompt’s count too, and the cost only when aider printed one. It reports no context window and no rate limits. | No trust step; the glow is on from the start. It is only ever green: aider tells MaeVae when a turn is done, never that it waits for an approval. | No |
| Grok | Its own sign-in inside the pane, or an xAI API key set in the profile. This is xAI’s Grok, not Groq. | model · tokens · context · time · cost quota n/aLines changed too. Grok reports no rate limits. | No trust step: close and reopen the pane, and the glow is live. | No |
| Plain shell | None. | no figuresMaeVae reads nothing for a shell pane. | It does not light up: the attention queue covers the six agents, not a plain shell. | No |
Limit. The figures are read from each agent’s own files on disk, and Claude Code’s quota from Anthropic’s usage endpoint; they are not a bill. A figure an agent does not report is grayed in the app, and its tooltip says why.
Three agents, three strips
One window with a Claude Code, a Codex and a Gemini CLI pane. Each pane’s header and the strip under its terminal fill as its agent reports, and a figure the agent does not report, grayed in the app, is rendered here as n/a.
Client work runs Claude Code: its header shows opus and, on a wide screen, $1.86, 184.0k tokens, 5h 31%, wk 46% and a context gauge at 41%; its strip reads Model opus, Cost $1.86, Weekly 46.0%, Context 82.0k of 200.0k, 41%. Side project runs Codex CLI: its header shows gpt-codex and, on a wide screen, 96.0k tokens, 5h 12%, wk 22% and a gauge at 22%; its strip reads Model gpt-codex, Cost n/a, Weekly 22.0%, Context 60.0k of 272.0k, 22%. Research runs Gemini CLI: its header shows gemini-pro and, on a wide screen, 140.0k tokens and a gauge at 5%; its strip reads Model gemini-pro, Cost n/a, Weekly n/a, Context 52.0k of 1.0M, 5%. Each header leaves out the figures its agent does not report. The names, lines and figures are examples, with more example lines and strip figures on a wide screen. Where motion is allowed, a short rendered scene plays once and ends on this frame: the headers and the strips fill one after another, their figures count up, and the figures an agent does not report arrive as n/a.
Local and other models
The agent Grok is xAI’s Grok, not Groq.
Codex, aider, Qwen Code and Grok profiles have an Endpoint field. Point it at a local model server on your own PC, or at an API provider you choose, such as Groq or OpenRouter.
At launch, a profile’s Endpoint section also takes a pasted key or address: paste one, and MaeVae fills in the profile.
In a Codex, aider or Qwen Code profile, List models asks that server which models it serves, only when you click it, or, at launch, when you paste an address into the Endpoint section. A Grok profile has an Endpoint but no Model field and no List models.
MaeVae still drives the agent’s own command-line tool; the profile’s endpoint and model settings decide which model it uses. Qwen Code uses an endpoint only when the key, the endpoint and the model are all set.
Guide: Other models
Questions about this page
The Support page says what to put in a message.
Write to support@maevae.com. The Terms page sets out the license.
MaeVae is coming to the Microsoft Store. The Store listing is not public yet; the Store button already points to it.