Chat & models
The chat is the whole product. Here's the interface, the models, and the input row controls.
The chat top bar
From left to right across the top of the chat app:
- Micro Learner logo — back to the home page
- Home — home page
- New chat (+) — start a fresh conversation
- Share — share the current conversation as a link
- Search (magnifier) — find a past conversation by text
- Settings (gear) — the settings modal
- Report a bug — opens the bug-report form
- More (⋮ desktop / ≡ mobile) — overflow menu: Home page, Share conversation, Report a bug (mobile also: New chat, Search conversations, Settings)
The input row
At the bottom of the chat, the message box has a set of controls:
- Paperclip — attach files or images, or open the MCP connectors menu
- Search toggle — turn web search on/off for a message (Google via Serper; subject to your plan's monthly quota)
- Context+ toggle — enhanced context / skills mode when available
- Model picker (right side) — choose the model for the conversation
- Send button
Models
All chat runs on DeepSeek (api.deepseek.com), currently the deepseek-v4-flash family. The picker shows user-facing names:
| Name | What it's for | Availability |
|---|---|---|
| Instant | Default, fastest — everyday builds and questions | All plans |
| Expert | Deeper reasoning for harder problems | All plans |
| DeepSeek v4 Pro | Frontier-class reasoning, full context | Max plan and above |
Attachments
- Images are described by a vision model so the AI knows what they contain.
- Office files (xlsx, docx, etc.) are read server-side and folded into context — you can say "summarize this sheet" or "fix this budget."
- Generated outputs are attached as artifact cards (see Files & artifacts).
Clarification cards
When a request is ambiguous, the AI can ask a multiple-choice question inline — single, multi-select, or ranking. For example, "make a presentation about our product" might prompt: "Which audience? Investors / Customers / New hires." Answer the card and it continues with that context baked in.
Live status bar
While the harness works, a status bar shows pipeline stages — memory/context → AI processing → response building. When a run is executing code server-side, this is where you can see it working; a stop control appears so you can cancel a long run.
Conversation tips
- Ask for files, not explanations — that's what the harness is for.
- Iterate: "now add a second sheet…", "change the colors…", "make it weekly instead of monthly."
- Ask it to save the workflow as a skill when you find one you'll repeat (see Skills).
- Conversations persist per user and archive after ~14 days idle — they rehydrate when you return.