TL;DR: AdaL can now tell you what people are saying online: ask "What is trending in AI this week?" and it looks across social sites for the answer. Compaction was rebuilt to retain more of a long conversation, and /compact takes direction. Browser use supports Microsoft Edge, GPT-6 Sol and GPT-6 Luna are in /model, and Jev nudges a model that stops before the job is done.
Ask what people are saying
Some questions are not answered by your codebase or the docs. They are answered by whatever people posted this week. What is trending in your field? How did your launch land? What are users saying about a library before you adopt it?
Now you can ask AdaL in plain words:
What are people saying on X about Claude Code this week?
AdaL looks across social sites, reads what people posted, and answers in the chat.
In the demo above, AdaL drives a logged-in browser through X search, then summarizes the five most talked-about posts with links.
Use cases
- Read the room before a launch. "What are people saying about agent CLIs?"
- Hear from your users. "What are people saying about
<your product>?" - Pick a library. "What are people saying about Bun versus Node?"
- Stay current. "What is trending in AI this week?"
Better compaction
A long session fills AdaL's memory. When that happens, AdaL summarizes the conversation and keeps going from the summary. We rebuilt that summarization so it retains more information.
Tell /compact what to keep
You can now give /compact direction:
/compact keep the API design details
AdaL keeps those details when it condenses the conversation. A bare /compact still works. Add guidance when the next step depends on particular decisions.
Reach for it when:
- You finish one part and start the next.
/compact keep the files we changed and the failing test - You are deep in a debugging session.
/compact keep the error and what we ruled out - You settled a design.
/compact keep the API design details
See the slash-command docs for /compact and the other session commands.
New models
- Claude Sonnet 5.5. Anthropic's newest Sonnet, with a 1M-token context window. Sonnet 5 stays the default; pick 5.5 from
/model. - GPT-6 Sol and GPT-6 Luna. New OpenAI models, both in
/model. - GPT Image 2.5 Flare and Sunburst. OpenAI's latest image generation, for photorealism, text inside images, and precise edits.
Browser use works in your Edge
AdaL's browser use now finds Microsoft Edge, alongside Chrome, and can bring your existing Edge profiles along. It works where you are already signed in, so a task like "open my X profile and tell me what my latest post says" just runs.
More in the browser use docs.
Also new
- Models that stop early get a nudge. When a model stops awkwardly before finishing the task, Jev detects it and nudges the agent to carry on.
Try it now. Update AdaL and ask it "What is trending in AI this week?"
Full notes are in the changelog.