Skip to content
Download Docs
Docs / Products / MaxiMind Chat

MaxiMind Chat

Conversations with Claude, Codex and Antigravity in the browser, saved to your account.

Open it

https://maximind.dev/chat — sign in with your MaxiMind Google account. Access is granted per person by the owner (/admin/chat); the owner account is always on. Until it is enabled you see a short notice instead of the app.

What it does

  • Conversations are saved on the server under your account: list, search, rename, delete, export as Markdown.
  • Pick a model per conversation, grouped Claude · Codex · Antigravity · Auto, from the same live catalogue as the API. Models the owner has denied are never offered.
  • Every answer shows which model produced it and how long it took; when the server had to switch models, the answer says so.
  • Share a conversation as a read-only link (/share/<token>); stop sharing at any time. Deleting a conversation ends its share link.
  • Light and dark appearance, English and Vietnamese interface, phone-friendly layout, keyboard shortcuts (⌘N new, ⌘K search, Esc close).
  • Settings shows your own use in the last 30 days and links to your MaxiMind Code computers.

Tools

The model can use MaxiMind's own services when a request needs them, and each use is shown as a step above the answer:

Tool What it does
youtube_transcript transcript of a YouTube/TikTok/… video (captions when available, otherwise generated)
video_info title, channel, duration, views of a video URL
web_read main text of a public web page as Markdown
transcribe_url Whisper transcription of an audio/video file at a public URL (minutes)
speak text to speech with a MaxiMind voice; the answer gets an audio player

Tools are read-only wrappers over the server's services, run with your session (never a key), refuse local/private addresses, and are limited to four rounds per message. Each round is one model turn and is metered like a message.

Skills

The Skill menu lists the approved skills from MaxiMind's library (the same ones reviewed at /admin). Choosing one makes its SKILL.md the instructions for that conversation and attaches its small reference files. Owners curate the library; a skill folder that a user submits becomes selectable only after approval.

What it does not do (yet)

  • No file uploads or attachments; no browsing beyond the tools above.
  • Answers arrive in one piece with a live "thinking… n s" indicator, because the engines' answers are buffered by the server.
  • No memory across conversations, no analysis of conversations.

Feedback and privacy

Under each answer you can mark it good or not helpful (with an optional note), and after an answer the page asks once "Did this help?". Conversations are saved to your account, and the MaxiMind owner may read them — together with your thumbs and an automatic outcome label — to improve the service. Delete a conversation to remove it from that review as well.

Limits

Up to 12 of your requests run at once (shared with MaxiMind Code and the API); a message is at most 60 000 characters; a conversation longer than the model's window must be exported and restarted. Every message uses shared model capacity and is metered under your account with surface: "chat".

API (browser session only)

Method Path Purpose
GET /api/chat/me account, grant, model list
GET / POST /api/chat/conversations list (with ?q=) / create {model}
GET / PATCH / DELETE /api/chat/conversations/{id} read / rename or change model / delete
POST /api/chat/conversations/{id}/messages send {content, model}; answers as server-sent events (status, then done or error)
GET /api/chat/conversations/{id}/export Markdown download
POST / DELETE /api/chat/conversations/{id}/share create / remove the read-only link
GET /api/chat/usage your use, last 30 days
POST /api/chat/conversations/{id}/messages/{mid}/feedback thumbs on one answer: {"rating": 1 | -1 | 0, "comment": "…"}
POST /api/chat/conversations/{id}/outcome your answer to "Did this help?": {"outcome": "yes" | "partly" | "no" | null}

These accept only the signed-in browser session and same-origin requests; API keys and Code device credentials are refused.