Hirun Nemitha

We just shipped a new attribution protocol that fixes how we match citations to optimizations.

The old method used single signals (like "most recent") and created false positives. v2.6 requires multiple signals to agree before assigning attribution, which gives you accurate data on which strategies actually drive citations.

In the past 4 months, we ve shipped dozens of updates. From low-level protocol adaptations to top-level UI overhauls, every line of code has been rewritten with one goal: to build a smarter, faster, and more secure AI assistant for you.

Deep Intelligence & Reasoning

• Visible Thinking: We now support streaming reasoning processes for most major model providers. Watch your AI s thought logic unfold in real-time.

• Multimodal Power: Upload documents and capture screenshots directly. Give your LLM more context to work with &

• Expanded Protocols: Added native support for OpenAI Responses API and Google API for enhanced compatibility.

Developers don t hate automation.
They hate noise.

Most AI tools fail not because they miss bugs, but because they overwhelm teams with comments no one trusts.

That insight changed how we built PRFlow.

We optimized for: