Skip to main content

Threads

Threads is the conversational AI of MINEO. Chat with any model over your data, files, databases, and connected tools — directly inside your Projects, with full history, search, and organization.

Every Thread runs on an Assistant, and it is model-agnostic: the Assistant decides which language model answers, what data and tools it can reach, and how it behaves, but you can switch the model per conversation without losing context — no vendor lock-in. Every interaction stays governed: permissions, traceability, and per-message cost tracking apply throughout. To configure that behavior, see Assistants; this page covers using Threads day to day.

The Threads list​

Each Project has a dedicated Threads section accessible from the left sidebar. It shows all your conversations and lets you manage them from a single place.

From the list you can:

  • Search threads by name using the search bar.
  • Filter between three views: All, Pinned, and Archived.
  • Filter by Assistant or model — two dropdowns narrow the list by the conversation's Assistant or LLM model. They only list values that actually have conversations, each with its conversation count. A thread matches a model when it is the thread's per-thread selection, or the Assistant's default when the thread has no override.
  • Open any thread to continue the conversation.
  • Create a new thread with the New Thread button.
  • Delete all threads in the project at once, from the ⋯ menu next to New Thread.

Each item in the list shows the thread name, optional description, message count, total token usage (↑ input · ↓ output; hover to see the exact figures), and last activity time. A context menu (⋯) on each item gives you quick access to Rename, Pin/Unpin, Archive/Unarchive, and Delete.

Creating a Thread from the sidebar​

You can start a new conversation directly from the project sidebar without going to the Threads list. Hover the Threads header and click the "+" that appears beside it, or press Ctrl+Shift+O (Cmd+Shift+O on macOS) from anywhere in the project. The sidebar shows:

  • The available Assistants for the project — click one to open a new thread pre-selected with that Assistant. If you can edit an Assistant, its ⋯ menu offers Move up and Move down to arrange the list; the order is remembered in this browser.
  • Recent threads — click any to jump straight into the conversation.
  • A New link to open an empty new-thread view.
  • An ALL link to go to the full Threads list.
  • A small progress bar next to Threads with how much of your AI budget you have used; hover it for the breakdown by today, this week and this month. See Your personal view.

The Thread view​

When you open or create a thread, you see the chat interface:

  • Header — shows the thread name, with a pencil to rename it, and a back button to return to the Threads list. On existing threads the ⋯ menu starts a new thread, renames, shares or deletes it.
  • Message history — the full conversation displayed in order, streamed in real time as the AI responds.
  • Starters — on a brand-new thread, the Assistant's conversation starters are displayed as quick-start buttons. Clicking one sends that prompt automatically.
  • Assistant selector — on a new thread, a dropdown lets you choose which Assistant will handle the conversation. Type to search; Assistants shared by the organization carry an Organization tag. Assistants you are allowed to edit show a gear icon that opens their configuration in a new tab, and members (not viewers) see a New Assistant action at the bottom of the list: it opens the creation form and, on save, takes you to the new Assistant's configuration. Viewers only see the list.
  • Message input — type your message and press Send. You can stop a running response at any time with the Stop button.
  • Voice input — the microphone button records your voice (disabled, with the reason in its tooltip, when the organization has no dictation model; see Use only your own API keys); while recording, the bars next to the timer move with the level the microphone picks up. Stop the recording to transcribe it into the message input.
  • Attachments — the paperclip button attaches files to your message (up to 10 per thread). Media files (images, audio, video) can weigh up to 21 MB; documents (PDF, text, Office) are limited to 10 MB, since their content travels with the conversation. Uploading the same file twice to a thread reuses the existing attachment.
Spreadsheets and big files are processed with code

If the Assistant has code execution on an OpenAI or Anthropic model, spreadsheets and CSV files (any size) and documents over 1 MB are not read into the conversation: the Assistant opens them in its Python sandbox and answers with the results of its analysis. A blue notice above the message box tells you when this applies. Ask for totals, filters, comparisons or charts rather than for the file's content, and keep asking about the same file in later messages — it stays available for the whole thread.

Office files without code execution depend on the model provider

Word and PowerPoint (.docx, .pptx) work with any model: MINEO extracts their text at upload and sends that to the model. Spreadsheets and legacy formats (.xlsx, .xls, .doc, .ppt, .odt, …) are read natively by OpenAI models (including a spreadsheet summary of up to 1,000 rows per sheet); other providers cannot read them — if the selected model can't read an attached file, a warning appears above the composer and the message cannot be sent: switch to an OpenAI model, remove the file, or use an Assistant with code execution.

  • Initial form — if the chosen Assistant's input mode is set to Form, a brand-new thread shows that form instead of the free-text input. Fill in the fields and press the send button to start the conversation; the composer's typing, attach, and tool controls stay disabled until the form is submitted.
tip

Pinned threads appear at the top of the sidebar with a pink pin icon, making it easy to keep important conversations accessible.

Responses keep running in the background​

A reply does not depend on your browser tab. Once you send a message, MINEO keeps working on it on the server even if you close the tab, navigate elsewhere, or lose the connection.

  • While you are away, the run continues: tools are called, the answer is written and saved, and it is billed like any other reply. It stops on its own at the usual limits (the per-run request and token ceilings, or a 15-minute wall clock).
  • While it runs, the thread shows a spinner in the sidebar, in every tab where you have the project open, and next to the thread's title. Its ⋯ menu in the sidebar offers Stop, so you can end it from wherever you are.
  • When it finishes, the thread shows a blue dot in the sidebar and in the Threads list, wherever you are in the project. Opening the thread clears it. The dot only appears for threads you are not looking at: a reply that ends while the conversation is on screen never marks it.
  • If you open the thread while the reply is still being written, you see the progress so far immediately and the rest keeps streaming live, exactly as if you had never left. The Stop button works there too and cancels the run on the server.
  • Closing the tab is not Stop. To cancel a reply, press Stop (in the thread or in its sidebar menu): only that ends the run. Pending tool approvals reached while you were away wait for you in the thread.
info

If the server restarts while a reply is in progress, the run is lost: within a couple of minutes the message is marked as failed with an error bubble, and you can send it again.

Commands​

Type / on an empty message box of an existing thread to open the command palette. Pick a command with the arrow keys and Enter, or click it. Commands are actions run by MINEO, not text: nothing is sent to the model, and the message box is cleared. They are not available on a brand-new thread until the first message has been sent.

CommandWhat it does
/newOpens a new conversation with the same Assistant.
/compactSummarizes the conversation to free space in the context window. See Context window and compaction.

Context window and compaction​

Every model reads the conversation through a context window of limited size. The thread header shows a small progress bar with how full the space the conversation can use is (the window minus what the model reserves for its reply); hover it for the token count. The bar appears once the first reply has arrived and only for models whose window MINEO knows, and it reflects what the model read on the last reply. At 70% of that space the conversation counts as long (MINEO compacts it by itself at 85%, so the warning always comes first): the bar turns amber with a warning badge and a strip above the message box offers New conversation (a fresh thread with the same Assistant, cheaper and faster for a different topic) or Compact. The strip can be closed for that thread; the same two actions stay in the bar's popover.

The popover also breaks the window down by category, so you can see what is filling it: system prompt, system tools, MCP tools, skills, attachments, compaction summary, messages, and free space. A stacked bar shows the split at a glance, and clicking a category unfolds what it is made of, heaviest first:

  • System prompt: the base prompt, your custom instructions, and the Assistant's instructions.
  • System tools: the tool instructions and each tool's definition.
  • MCP tools and Skills: one line per server or skill.
  • Attachments: the files sent whole on the last reply, by name.
  • Messages: your messages, the assistant's replies, and tool results.

These figures are estimates. MINEO tokenizes what it sent on the last reply and reconciles the result with the count the provider reported. Files the provider reads on its own — attached to the thread or to the Assistant — are not tokenized locally: Attachments is whatever the provider counted beyond the measured text, so it carries no per-file figure. Skills the model loads on demand are not counted until it loads one; from then on their text counts as a tool result. Hover the info icon next to the total for the model that measured the window. A category only appears when it is non-empty, and the breakdown shows up from the next reply on threads that predate it.

To free space, compact the conversation: press Compact in that popover, or type the /compact command in the message box. MINEO condenses the older part of what the model receives into a summary and keeps the recent messages whole. The chat keeps every message: nothing disappears from the thread, and a "Conversation summarized" card marks the point where it happened. The model recovers details that fell out of the summary on demand, by reading the conversation archive.

  • It needs at least three messages; below that there is nothing to compact.
  • A pending tool approval is left untouched, so you can still resolve it afterwards.
  • The summarizing call is billed to you like any other reply and counts against your AI limits. It also obeys the per-response token ceiling, so a very long thread may be refused as too large to compact.
  • The thread's token totals are kept.
  • Compaction also happens on its own when a reply would not fit: the reply then starts with the same card. Old tool results are cleared before anything is summarized, and repeated reads of the same file keep only the latest.

Organization administrators choose which model writes the summary in Organization → AI → Models → Default models — platform and BYOK models are available. When none is set, the thread's own model is used; a fast, low-cost model is recommended.

info

Responses are streamed in real time. Each assistant reply shows its token usage (↑ input · ↓ output) in the message footer — for tool-only runs, under the task block — and the Threads list shows each conversation's total. Hover the figure for the reply details: the model that answered, when it replied, the response time and the time to the first token, how many model requests, tool calls and native web searches the turn took, input tokens (with the part read from or written to the provider's cache), output tokens (with the part spent reasoning), the context the model held on its last request, and — when the model has a known price — the estimated cost, its input/output split, what the web searches cost (billed per search, apart from tokens) and what the cache saved. A link opens your consumption page. To cap and monitor spend across the organization, see AI Cost Limits.

See also​

  • Assistants — configure the AI agents that power your threads: model, instructions, capabilities, knowledge, and tools.
  • Work mode — run an Assistant as a long-running agent with its own workspace.
  • Knowledge bases — the personal memory and team knowledge an Assistant reads and writes during a conversation.
  • Custom instructions — personal context injected into every Assistant you chat with.
  • Conversation starters — the suggested prompts shown when you open a new thread.
  • AI Cost Limits — track and cap AI spend per organization and project.