Field manual · SQN-01
Flying Squadron
Everything you need for day-to-day use, in about five minutes. If something here doesn't match what you see on screen, tell us: we fix the manual first.
01 · Boarding
Sign in
- Open chat.jmni.ai and enter the email address you were invited with.
- Cloudflare emails you a one-time code. Enter it and you're in; your first sign-in creates your account.
- You stay signed in for seven days on that device.
No code after a minute? Check your spam folder and make sure you typed the exact address you were invited with. If it still doesn't arrive, that address isn't on the manifest yet: ask for access.
Put it on your phone
iPhone: open chat.jmni.ai in Safari, tap Share, then Add to Home Screen. Android: open it in Chrome, open the menu, then Add to Home screen. It launches full-screen as Squadron, with its own icon.
02 · Conversations
Talk to it like a colleague
Type in the composer and press Enter to send (Shift + Enter for a new line). The start screen offers a few openers to get going: Brief me, Scan the web, Plot a course, Draft a transmission.
- Every chat is saved in the sidebar. Search them, and pin, rename or file them into folders from each chat's ⋯ menu.
- Change course by editing a message you sent or regenerating an answer. Follow-up suggestions appear under each answer.
- Off the record: the dashed-circle button at the top right starts a temporary chat, which is not saved to your history.
- It reasons before it answers, so a hard question can take a few seconds before text starts to stream. With an assistant you can turn that off for faster replies (next section).
03 · Assistants
Pick an assistant for the job
The model menu in the composer lists ready-made assistants next to the general models,
deepseek-v4.1-flash and glm-5.3-flash. Those two share the fleet and take turns: pick
either, and whichever is running at the moment answers. Each one has its own
instructions and starter prompts on the start screen; all of them run on the same fleet and share your limits.
| Assistant | Good for |
|---|---|
| Research | Questions that need the live web. Searches first and cites a source for each claim. |
| Writer | Drafting and tightening emails, notes and posts in your voice. Fast by default. |
| Coder | Writing, explaining and fixing code. |
| Tutor | Learning something step by step. Checks your understanding and won't just hand over homework answers. |
| Analyst | Numbers, spreadsheets and charts. Calculates instead of estimating (see Charts and numbers). |
Thinking: on or off
The sliders button next to the assistant's name opens Thinking. On lets the model reason before it answers, which is better for hard questions; Off (faster) starts the reply straight away. Each assistant has a sensible default, and your choice applies to that chat.
04 · Documents
Bring your documents
Click + then Upload Files, or drag files onto the window. PDFs, Word files, Markdown, plain text, CSV and code all work: up to 25 MB each, ten at a time. The same menu can Attach Webpage (Squadron reads a page from its URL) or Reference Chats (bring in an earlier conversation).
Two ways to read a file
| Mode | What happens | Use it for |
|---|---|---|
| Focused retrieval | The default. Squadron finds the passages that match your question and cites them. | Looking things up in long files, or across many |
| Entire document | The whole file goes to the model with your message. | Summaries, reviews and "what's missing" questions on a shorter file |
To switch, click the file in the composer and flip Using Focused Retrieval. Answers that drew on your files carry numbered citations, and Sources under the answer shows the exact passages.
Long books stay on focused retrieval. In entire-document mode the whole file is re-read with every message in that chat, which draws on your monthly allowance (see Limits).
Keep a set of files together: knowledge bases
For documents you come back to (a handbook, a project's papers, a course's notes), make a knowledge base once instead of uploading the files to every chat: Workspace → Knowledge → Create, then add files. In any chat, type # in the composer and pick it; Squadron looks things up across all of its files and cites them. Your knowledge bases are private to you.
05 · Web search
Search the live web
Click Integrations, next to +, and turn on Web Search. Squadron runs the search, reads the top results and answers with sources. Use it for anything recent: news, prices, releases, documentation.
Searches go out through a search service we run. Search engines see the query and our server, not you, your address or your account.
06 · Charts and numbers
Charts, calculations and spreadsheets
Squadron can write a short Python program and run it to get an exact answer: totals, percentages, statistics, or a chart from your numbers or an uploaded CSV or Excel file. The Analyst assistant does this by default; with any other, turn on Code Interpreter under Integrations. A code block in an answer also has a Run button.
The program runs in your own browser, not on our servers. The first run in a session takes a few seconds while it loads.
07 · Scheduled tasks
Have it do something on a schedule
Ask for a morning briefing, a weekly summary or a reminder with context, and Squadron runs it on its own: click your avatar, then Automations → Create. Give it the instructions, an assistant and a schedule. Each run lands as a new chat in your sidebar.
- Up to five scheduled tasks each, running at most once an hour.
- Every run is a normal message, so it counts toward your monthly allowance like one you sent.
- Pause or delete a task from the same page at any time.
08 · Memory
Teach it what to remember
Squadron can keep short facts about you (your role, your projects, how you like answers) and use them in every new chat. Say "Remember that I report in metric units", or add them yourself: click your avatar, then Settings → Personalization → Memory. That's also where you review, edit and delete everything it has kept.
09 · Limits
Limits
Squadron runs on a fixed fleet shared by a small crew, so each member works within these limits:
| Limit | Value |
|---|---|
| Pace | 30 requests a minute; one message uses two or three, since titles and follow-up suggestions count too |
| One message | About 1,000,000 tokens (roughly 750,000 words), the model's full context, counting the earlier conversation and any documents read in full |
| Monthly allowance | Roughly 50 million tokens read, or 12 million written |
| Uploads | 25 MB per file, ten files at a time |
| Scheduled tasks | Five each, each running at most once an hour |
If you reach one, Squadron tells you in plain words, including the date your allowance resets. The per-minute pace clears within a minute and the allowance resets on the 1st of each month.
Long chats spend the allowance fastest. Every message re-reads the whole conversation so far, so a chat holding 500,000 tokens uses about 1% of the monthly allowance per message. When a thread has grown long, start a fresh one (your memory comes with you) and keep large files on focused retrieval.
10 · Your data
Your data
Chats, files and memories are stored on hardware we own, and no third-party AI sees them. Delete any chat from the sidebar, or export or delete all of them under Settings → Data Controls. The operator administers the system and can access stored data; the details are in Privacy and Claims.
11 · Help
Help
Something broken, slow or confusing? Write to [email protected] with what you tried and roughly when. Outage alerts reach us automatically, but a note with what you saw still helps.
Building something? The API reference covers keys, models and embeddings.