New in 1.8.0

• Reasoning levels — Thinking is no longer just on/off. A level selector beside the switch
offers Default, Low, Medium and High. Some models (gpt-oss and others) ignore an on/off
request entirely and respond only to a level, so previously there was no way to influence
their reasoning at all.

• Ollama Cloud support — cloud-hosted models always run at their maximum context and ignore
any size you set, so the Context size control is now hidden for them and the usage gauge
measures against the model's real maximum instead.

• The context gauge has moved to the left of the message box, where it is easier to see while
typing.

Fixed

• A model's maximum context was remembered as "unknown" if the lookup happened to fail once —
for example with the Ollama server briefly unreachable — and was never retried, leaving the
Context size slider capped far too low from then on.

• Settings could be lost when two chat boxes saved at the same time: each wrote back its own
view of the list, discarding the other's changes.

• Regenerating a reply kept the previous answer's "Thought process" panel when the new reply
had no reasoning of its own.

• The gauge no longer states that Ollama's default context is 4096. Ollama sizes it by the
server's available VRAM (4K, 32K or 256K), so the tooltip now says so and recommends setting
a Context size if you want an exact figure.

• Token usage is cleared rather than left showing an old reading when Ollama reports no counts.

New in 1.7.0

• Per-model settings — a gear icon next to the model selector opens settings for the
selected model: enable Thinking for reasoning-capable models, override the sampling
Temperature, and set the Context size. Saved per model and shared instantly across
every tab and chat box.

• Reasoning visibility — models that think out loud now stream their reasoning into a
collapsible panel above the answer. It stays with the conversation, so you can expand
it again after a reload.

• Context window gauge — a colour-coded ring beside the model selector shows how much
of the context window the conversation is using, from the exact token counts Ollama
reports. Hover it for the numbers.

• Saved prompts now sync live between chat boxes.

Fixed

• The Thinking setting had no effect on models that reason by default (qwen3,
deepseek-r1) — they kept producing reasoning with it switched off.

• Area capture no longer drifts from the region you selected; the crop now accounts for
display scaling and browser zoom.

• Capture silently returning nothing after navigating to a new page or reloading the
extension, and failing outright on sites whose content security policy blocked it.

• The chat log no longer stops following fast responses part-way through.

• The chat box no longer shakes when dragged to full window height.

• A finished reply could come back as an empty message with a spinner after refreshing
the page, and deleted messages could reappear.

• A dropped connection left the spinner running forever with no explanation; failures
now show an error.

Note: Thinking and the reasoning panel require a recent Ollama version and a model that
supports it. Raising Context size beyond what your hardware can hold will push the model
onto the CPU and slow generation considerably.

code fixes

packages bump + small bug fixes

bug fixes

fixed bug issues related to the chat input element

Added files attachment functionality.