class

ion7.llm.chat.Thinking

_in_think boolean

_think_text string[] Accumulator for the active think block.

_open_tail string Last 6 chars seen outside a think block.

_close_tail string Last 7 chars seen inside a think block.

_tok_count integer Tokens emitted inside the active block.

_all_think string[] Concatenation of every closed block.

Functions

Thinking.new Thinking:reset Thinking:in_think Thinking:active_token_count Thinking:thinking ? Thinking:force_close

Thinking.new

Thinking.new()

→ ion7.llm.chat.Thinking

Thinking:reset

Reset to the initial state — call between generations. Returns the thinking text accumulated since the last reset (useful when the upstream loop wants to clear the buffer AND keep the trace).

Thinking:reset()

→ string

Thinking:in_think

True when the buffer is currently inside a `` block.

Thinking:in_think()

→ boolean

Thinking:active_token_count

Total tokens consumed inside the active block. Resets to 0 each time a block closes. Used by the reasoning-budget guard.

Thinking:active_token_count()

→ integer

Thinking:thinking

Concatenated text of every closed `` block since the last `:reset`, plus whatever is currently buffered in the active one.

Thinking:thinking()

→ string?

?

Feed a decoded piece. Returns one of the three transitions : - `"content"` — emit `text` on the assistant channel. - `"thinking"`— emit `text` on the reasoning channel. - `"split"` — the piece straddles the `` open tag : `prefix` is content, `suffix` is thinking. Same for `` in the other direction.

?()

piecestring

→ stringkind `"content" | "thinking" | "split"`

→ stringtext or prefix.

→ string?suffix on a `"split"` transition.

Thinking:force_close

Force-close the active think block. Used by the reasoning-budget guard when the model exceeds its allotment without emitting ``. Returns the body collected so far.

Thinking:force_close()

→ string