Think

AI reasoning display.

---

Purpose

Show the AI's "thinking" — chain-of-thought, reasoning, internal notes. Can be hidden or visible.

Use <think> when the application needs an explicit reasoning container that can be hidden, revealed, or audited separately from the user-facing response stream.

---

Try it Live

editor.g

12345678910111213141516171819

Grain document editor

Edit the Grain document and inspect the rendered preview beside it.

<message role="assistant"> <think model="chain-of-thought" visible="false"> The user wants a weather update for Mumbai. </think> <stream>Checking Mumbai weather...</stream> </message> <tool name="get_weather" args='{"city":"Mumbai"}' status="running"> <input>City: Mumbai</input> <progress value="45" max="100">Fetching live conditions</progress> </tool> <tool name="get_weather" status="complete"> <result temperature="28" condition="sunny" humidity="65" /> </tool> <message role="assistant"> <stream>It is 28°C and sunny in Mumbai.</stream> </message>

preview

Interactive preview initializes after hydration.

Interactive preview initializes after hydration. Open the docs preview at /grain/; the site is mounted under the project base path.

---

G-Lang

<think model="chain-of-thought" visible="false" depth="medium">
  The user is asking about weather. I should call the weather API.
</think>

---

Attributes

Attribute	Type	Default	Description
model	string	chain-of-thought	Reasoning model
visible	true | false	false	Default visibility
depth	shallow | medium | deep	medium	Reasoning detail

---

States

HIDDEN ↔ VISIBLE
   ↕
EXPANDING/COLLAPSING

---

Events

Event	Description
think.reveal	User revealed thinking
think.hide	User hid thinking

---

Examples

Hidden by default

<think visible="false">
  User intent: weather query, location: Mumbai
</think>

Visible (for educational/debugging)

<think visible="true">
  Let me solve this step by step:
  1. First, identify the variables...
  2. Then, apply the formula...
</think>

Tree-of-thought

<think model="tree-of-thought" visible="true" depth="deep">
  Considering multiple approaches:
  
  Branch A: Recursive solution
    - Pros: Elegant, easy to understand
    - Cons: Stack overflow risk
  
  Branch B: Iterative solution
    - Pros: Memory efficient
    - Cons: More verbose
  
  Decision: Branch B for production use.
</think>

---

Related

• Stream — Output text
• Tool — Tool execution
• Playground — Try it live

Contract Notes

• Treat <think> as intentional structure, not a dumping ground for arbitrary hidden text.
• Keep visibility explicit so adapters do not guess whether reasoning should be shown.
• If the reasoning content is not meant to be surfaced at all, omit it rather than depending on renderer quirks.