What's new in tl v0.7.12: no more ceiling on deep data sweeps
Two smooth upgrades land today: your AI agent can now walk through huge result sets end-to-end — past the old 10,000-row ceiling — when answering broad sweeping questions, and setting tl up inside the Claude desktop app just got a lot more reliable.
A new release of the tl CLI is out today: v0.7.12. It’s a focused update — most of what changed sits one layer below where you usually look, in the engine that powers the deeper data questions your AI agent asks on your behalf. As always, your agent (Claude Code, Gemini, Codex, or OpenCode) picks all of this up automatically — there’s nothing new to learn.
No more ceiling on “show me everyone” questions
Some questions are naturally huge. “Pull every video that’s ever mentioned sustainability.” “Walk through every brand mention we’ve ever indexed for this category.” “Give me the full list, not just the top few thousand.” These broad sweeps used to hit an invisible ceiling around the 10,000-row mark — your agent could pull the strongest matches, but the long tail past that ceiling was out of reach without falling back to slower, more expensive workarounds.
That ceiling is gone. Your agent can now walk the full result set end-to-end, page by page, at a flat per-page cost no matter how deep it goes. Whether the answer is a hundred rows or a hundred thousand, the trip down stays smooth.
What this unlocks, in plain English:
- “Every video on the platform that’s ever mentioned a competitor of ours — pull the whole list, I want to see how deep the bench is.”
- “Sweep through all the videos in the cooking category from the last year, I’ll narrow down once I see the spread.”
- “How many videos in total touch ‘electric vehicles’ at all? Walk the full set so I get an exact number.”
You’ll mostly notice this when you ask your agent a question that used to come back with a quiet caveat (“I can only see the first few thousand”) — that caveat is gone now. Big questions get full answers.
A small note on cost: walking deeper still costs more credits, because each page returned is real data. But the per-page rate stays flat, which makes large sweeps predictable to budget. Ask your agent to preview the cost before running a wide one — same way you would for any custom query.
Smoother setup inside the Claude desktop app
If you use Claude as a desktop app rather than in the terminal, tl setup claude is now a lot more reliable. It finds the bundled Claude binary that ships with the desktop app automatically and installs tl as a proper plugin there — the same way it already did for terminal users.
If you’ve already set it up, the next run will quietly tidy up any leftover copies of the standalone version that exactly match the bundled ones, without ever touching files you’ve edited yourself.
Nothing to do on your end except, if you haven’t yet, run tl setup claude once.
Updating
Run tl update to pick all of this up — or just keep working, and the auto-updater will catch you up on your next command. tl changelog always shows the running log of what’s landed.
Happy hunting.