Start Your First Chat

Once a provider is configured, sending the first message is just one tap. But the chat screen hides a lot of optional power: attachments, tool switches, thinking toggles, model switching, session search, export. This page gets the minimum loop working, then tells you what each control does and where to find it.

Read This First

Send Your First Message

Tap the Chat tab. If you've followed along, the screen should show:

• Top bar: hamburger icon top-left (opens the session list) / center shows "New Conversation" / top-right may show model controls
• A big empty middle area (the chat surface)
• Bottom: input field with placeholder Message, a plus button on the left (attachments), and a send button on the right

If the title says "Select a model to start"

You haven't picked a Current Model. Go back to Settings → Current Model → Model and choose one.

Easiest first message: type "Hello", tap send.

If everything is wired up:

1. Your message appears as a bubble on the right.
2. An AI bubble appears on the left, with text streaming in word by word.
3. After streaming finishes, small metrics (duration, token count) may appear under the bubble depending on your settings.

When you see the streaming reply, you're done.

When It Doesn't Work

Check in this order:

Symptom	Real Cause	Fix
Title says "Select a model to start"	No Current Model	Settings → Current Model → Model
Send button is grayed out	Empty input / no model	Check both
Alert "Authentication failed"	API key wrong or expired	Go back to Add Your First Provider and verify
Alert "Connection failed" / timeout	Network can't reach the vendor	Enable per-provider proxy; or change network
Alert "Model not found"	Model ID wrong, or key lacks access	Use Fetch from cloud to re-pull the model list
Reply gets cut off mid-stream	Vendor rate limit / context overflow	Wait and retry; or lower Max History Messages in Preferences

What's Around the Input Bar

From left to right, top to bottom:

Element	Looks Like	Does
Plus (attachments)	+ to the left of the field	Opens the attachment menu: photo, camera, voice, file
Input field	Rounded rect, placeholder Message	Type here. Multi-line supported — Return adds a new line, not send
Tool switches	A row of small icons above/below the field	Temporarily toggle tools for the next turn (see below)
Send button	Blue circular arrow on the right	Tap to send. Long-press for send-with-options (Advanced)

What's in the Plus Menu

The bottom sheet (ChatViewTelegramComposer) lists five options:

Option	System Icon	Does
Choose Photo	photo	Pick an image from your library as a multimodal input. Needs a vision-capable model.
Take Photo	camera	Trigger the system camera for a fresh shot.
Record Voice	waveform	Capture audio inline as an attachment.
Upload from Voice Memos	music.note.list	Pick from existing iOS Voice Memos.
Choose File	doc	Pick anything (PDF, TXT, JSON, …) from the Files app.

Model Must Support It

• Image attachments need a vision model (GPT-4o, Claude 3.5 Sonnet, Gemini, …).
• Audio attachments need a model that accepts audio (GPT-4o Audio, Gemini, …).

When the model doesn't support the modality, the attachment is dropped and the AI never sees it.

Switch Models

Tap the model selector in the top-right of the chat to see every enabled model. The switch applies only to the current conversation — other sessions keep their own selection.

To set the default model for new conversations: Settings → Current Model → Model.

See How the AI Thinks

For reasoning-capable models (GPT-5, Claude 4, DeepSeek R1, Gemini Thinking, …), a collapsible gray "Thinking" panel appears inside the AI bubble while the model reasons. It's the live draft of the model's reasoning.

• Auto-expand: Settings → Display & Experience → Visuals → Auto-Preview Thinking, toggle on
• Hide it: collapse the panel manually

Read Replies Out Loud

Long-press any AI bubble → Read Aloud. First time, configure a TTS model:

Settings → Conversation → Text-to-Speech (TTS) → pick a TTS model

Available TTS models come from your providers (OpenAI's tts-1 / tts-1-hd, or any OpenAI-compatible TTS endpoint).

Export the Whole Session or Just a Snippet

Every message's long-press menu has two export options:

Option	Does	When to use
Export Entire Session	Saves the full conversation	Archiving
Export Up to This Message (with context)	Saves everything up to (and including) this message	Sharing a snippet, or trimming context

Formats: PDF / Markdown / TXT. The share sheet hands them to Files or your favorite app.

Advanced

Tool Switches (Per-Turn)

The row of tool icons around the input field maps to:

• Web search (if the model supports it)
• MCP tools (if you've connected an MCP server)
• Skills (if you've imported any Agent Skill)
• Shortcuts (if you've bound any iOS Shortcuts)

These toggles affect only the next send. To enable/disable a tool category globally, use its full Settings entry: Tool Center, MCP Tool Integration, Agent Skills, Shortcut Tool Integration.

Multimodal: Send Images, Audio, Files

Flow:

1. Tap + → pick an attachment type.
2. A thumbnail (image) / waveform (audio) / file card appears above the input field.
3. To remove one, tap the × on its corner.
4. You can still type — text and attachments are sent together.
5. Tap send.

Multi-image: tap + → Choose Photo again — the picker supports multi-select.

Branching

To explore an alternate timeline from any past message, long-press it → Create Prompt Branch. A "Branch Options" dialog asks:

• Prompt only — copy the system prompt without messages
• Messages only — copy everything up to (and including) the selected message, including AI replies

The branch becomes an independent session; the original is untouched.

Search the Session List

Tap the hamburger to open the session list. The search field at the top (placeholder "Search session titles or messages") matches:

• Title fuzzy match
• Full-text message content — finds keywords inside any past message

Tapping a result jumps directly to the matching message ("message anchor").

Organize Sessions with Folders

Long-press a session → Move to Folder. Create folders from the session-list header. A session can live in one folder at a time; you can reorganize whenever.

History Window (Context Length)

The model only sees the most recent N turns, not your full history. Adjust it in:

Settings → Conversation → Preferences → Max History Messages

The default is balanced. Lower saves tokens but the AI forgets faster; higher lets the AI remember more but each request costs more.

When Nothing Else Works

• Verify the request actually went out: Settings → Extended Features → Debug shows the full outgoing URL, headers, and body.
• Temporarily switch providers: if one provider misbehaves, switch to another in Settings and retry.
• Reset the conversation: long-press the session → Delete, then start a fresh one.

Next

Chat is working → continue with the Interface Tour for a full map of the Settings page.

Or jump to whatever interests you:

• Chat & Models
• Tools & MCP
• Memory & Worldbook
• Daily Pulse