Add to Chrome
August 9, 2026

What's new in tl v0.9.7: a memory your agent keeps between sessions, linked reports that stay in sync, and a clearer whoami

Your AI agent can now remember the context you keep re-explaining every session — your accounts, your priorities, your naming conventions. Reports can pull entities from other reports as live linked sources instead of one-shot snapshots. And `tl whoami` finally tells you whether you're on the Industry Weekly newsletter.

> memory that persists, reports that link, whoami that tells you more # less repetition, more context — no client update required tl memory show read what your agent already knows tl memory add append a durable note in one call tl memory set rewrite the whole memory with care tl reports link pull entities from another report, live tl reports unlink detach a linked source tl whoami now surfaces Industry Weekly membership

A new release of the tl CLI is out today: v0.9.7. This one covers a few different tracks that landed together since v0.9.4: a durable memory your AI agent can read and write to remember context between sessions, linked report sources that stay in sync when the source changes, a friendlier tl whoami, and a pair of schema-doc fixes that quietly stop the most common raw-SQL wrong turns. As always, Claude Code, Gemini, Codex, and OpenCode all pick everything up on the next tl update.

A memory your agent keeps between sessions

Here’s the headline. If you’ve ever felt like you’re re-briefing your agent from scratch every morning — “remember, my top three verticals are finance, fitness, and food; my best guide brands are these; when I say ‘the mattress client’ I mean Nectar” — v0.9.7 gives you a place for all of that to live. Three new commands make it work:

  • tl memory show — read back everything your agent currently knows about you and how you work.
  • tl memory add "<note>" — append a durable note. Use it once per fact and let it accumulate. It also accepts --from-file for a longer paragraph you’d rather paste in from an editor.
  • tl memory set "<full text>" — rewrite the whole memory in one go. Handy for cleanups; refuses a blank input so you can’t wipe the memory by accident.

The memory travels with your account, so anywhere you use tl — same laptop, new laptop, a colleague’s shared session — your agent starts with the same context loaded.

What to put in it (this is where it earns its keep):

  • “My core verticals are finance, fitness, and CPG-food. I don’t work outreach on gaming.”
  • “When I say ‘the mattress client’ I mean Nectar. When I say ‘the coffee client’ I mean Trade.”
  • “I care most about renewal signals and views-per-subscriber. De-prioritise raw subscriber counts.”
  • “For any shortlist over 20 channels, always propose a Google Sheet export.”

Prompt shapes that start pulling on the memory naturally:

  • “Remember that my top three guide brands for the finance vertical are Personal Finance Club, Graham Stephan, and The Money Guy. Add that to your memory.”
  • “Before you shortlist for the coffee brief, check what you already know about my priorities and preferred output format.”
  • “Show me what you currently remember about how I work — I want to prune anything stale.”

The result: fewer opening paragraphs of context every session, and shortlists that arrive already shaped like the ones you asked for last time.

Linked report sources: reports that stay in sync

If you build reports in the ThoughtLeaders app, you know the two shapes an entity source can take today: a static list you paste or upload, and a filter that lives against the underlying data. v0.9.7 adds a third: linked sources, where one report includes (or excludes) the entities from another report — and the include set stays in sync as the source report changes.

Two new commands make this reachable from the CLI (and therefore from your agent):

  • tl reports link — attach another report as a live include or exclude source for the current report. Works for channels, brands, articles, and sponsorships.
  • tl reports unlink — detach a linked source when you no longer want it.

The practical difference vs. a bulk-import snapshot: if the source report grows, shrinks, or gets re-scoped, your dependent report picks up the change automatically. No more re-exporting and re-uploading every time the upstream shortlist gets updated.

Where this pays off:

  • “I have a report of every channel that ran a Nectar campaign in the last 12 months. Create a new report that excludes those channels from my active outreach shortlist, and link the two so the exclusion updates when I add new campaigns to the first one.”
  • “Take my ‘MSN channels over 500K subs’ report and link it as the include source on a new report that also filters by finance vertical. When the MSN roster grows, I want the finance cut to grow with it.”
  • “Unlink the old ‘Q1 renewal winners’ source from my current shortlist — I want a fresh Q2 base instead.”

If you’re the kind of user who maintains a handful of long-running reference lists and derives working shortlists from them, this is the release that stops the copy-paste treadmill.

tl whoami now tells you about the newsletter

Small quality-of-life fix: tl whoami (and the --md output) now surfaces your Industry Weekly newsletter membership right in the summary. If you’re subscribed, you’ll see it listed alongside your account details. If we couldn’t determine your membership one way or the other, nothing prints — which is a deliberate choice, because a misleading “not subscribed” would be worse than no row at all.

Nothing to do here — next time you or your agent asks “who am I logged in as?”, the answer is a bit more complete.

Fewer dead-ends when your agent explores the raw data

Two schema-doc fixes worth naming, both from real customer sessions where an AI agent wasted a round-trip on a name it thought existed:

  • The advertising-link url field is now correctly documented as urls (plural, and an array, since a single sponsorship can point at multiple destination URLs). The old singular name lingered in the schema doc long after the underlying rename; agents kept trying it and hitting a puzzling “unknown column”.
  • Channel audience size is now flagged sharply as subscribers when your agent is on the raw Postgres lane. The word reach is a legacy alias that only lives on the search-index side — mixing the two lanes was the specific pattern that kept tripping people up. The doc now calls out the difference in the same “these columns do NOT exist” style used for the other common traps.

You’ll feel this indirectly: a deep-dive research prompt that used to occasionally stall in the middle now just… doesn’t. Nothing about how you ask the question changes.

Updating

Run tl update to grab v0.9.7, or just keep working — the auto-updater catches you up on your next call. tl changelog shows the running log of what’s landed since your last release.

If you’ve been meaning to teach your agent your working style — your verticals, your guide brands, your naming shortcuts — this is the release to do it in one sitting.