For content operators
Read the world once.
Publish in your voice.
A daily briefing built from your sources, drafts shaped by your actual writing, and a clean path to publish across X, LinkedIn, and Substack — all in one place.
Closed beta. Sign in with an existing account.
The signal is real. The noise is louder.
Twenty open tabs. Six newsletters. Three Twitter lists. By the time you’ve caught up, the moment’s already passed — and most of what passed wasn’t worth catching.
Daily briefing
What actually moved overnight. Not who tweeted loudest.
Ledger watches your sources — X, LinkedIn, RSS, Substack, your inbox — and clusters the day's discourse into a handful of themes. Not grouped by author. Grouped by idea. So when seven different writers circle the same insight, that's one theme, not seven scroll moments.
- Two distinct sources before a theme surfaces — no single-poster echo chambers.
- Cross-references to your past insights so you see the through-line, not just today.
- Built nightly. Override with one click.
Tuesday, May 19, 2026
The morning briefing.
7 threads from 1,524 signals across your sources.
AI agent benchmarks miss the full picture of real-world capability.
Existing tests focus on retrieval or execution separately, leaving a gap in how we evaluate agents that must search and act in tandem.
Bret Taylor
“Introducing the latest agent benchmark from Sierra, τ-knowledge. Most benchmarks test either an agent's ability to find information…
Wed · ↗ article
Andrew Ng
“New course: Transformers in Practice. You'll get a practical view of how transformer-based LLMs work, so you can reason about their behavior…
Wed · ↗ article
Voice profile
Drafts that sound like you. Not “AI-written.”
Paste in a few paragraphs from your archive. Ledger extracts your sentence rhythm — how long, how often you open with a clause, your hedge ratio, em-dash density, the phrases you'd never say. Generation runs in two passes: a strategist surfaces claims, a ghostwriter renders them in your fingerprint.
- Reference accounts inform style. Anti-references inform what to avoid.
- Stance picker: agree + extend, civil disagreement, steelman + pivot, quote + take.
- Output reads like you wrote it, not like the assistant wrote it.
Cite source
Appends a “Source: @handle → url” line at the end
Agree + extend
Echo plus your value-add
Civil disagreement
Counter on the merits, not the messenger
Steelman + pivot
Strongest version of theirs → then your refinement
Quote + take
Surface without strong stance
Article distillation
Respond to the article. Not the hot take about it.
When someone posts about a piece on LinkedIn or X, the post is the discovery vector — not the substance. Ledger follows the link (yes, even lnkd.in interstitials), pulls the article via Readability, and grounds your draft in what was actually written. Citations land on the publisher, not the LinkedIn post.
- Reads through lnkd.in / t.co / bit.ly automatically.
- Article title, byline, and excerpt embedded in the draft for context.
- Falls back gracefully when the article is paywalled or unfetchable.
↗ Draft from briefing
Transformers in Practice
Transformer internals — attention, KV caching, quantization — are genuinely high-leverage skills most courses never connect to production. This one’s co-built with AMD, on AMD GPUs. That hardware intuition may not transfer cleanly to the CUDA environments most engineers actually ship on.
“You’ll get a practical view of how transformer-based LLMs work, so you can reason about their behavior, diagnose problems like slow inference, and make smarter decisions about deployment.”— Transformers in Practice · deeplearning.ai
↗ https://www.deeplearning.ai/courses/transformers-in-practice
One draft, many platforms
Write once. Ship to X, LinkedIn, or Substack.
Your canonical draft is a single document. Each platform tab — X thread, LinkedIn post, Substack essay — derives from it: respecting the 280-char ceiling, splitting at the right paragraph break, adding the right kind of citation, attaching the right kind of image.
- X thread split at sentence boundaries with per-tweet character counts.
- LinkedIn renders as a single post or a multi-slide carousel.
- Substack output is paste-ready Markdown.
Benchmarks test retrieval + action as a pair. That's not the hard part.
68/280The hard part is whether an agent holds state *across* the full handoff loop — retrieve, act, file back — without hallucination compounding at each step.
160/280No eval suite grades that yet.
30/280Source: Bret Taylor → https://www.sierra.ai/blog/tau-knowledge
62/280Visual toolkit
Quote cards. Charts. AI images. On-brand by default.
Set your workspace palette and fonts once. Every generated visual — quote graphics, LinkedIn carousels, stock charts, AI-generated headers — uses them. No generic AI aesthetic. No off-brand quotes.
- Five built-in plug-ins: quote cards, LinkedIn carousels, charts, watchlist sparklines, AI image headers.
- Polygon-powered market data. kie.ai for nano-banana and gpt-image-2 generation.
- Everything lands in your asset bin, ready to attach to any draft.
Add visual ▸
“
Quote card
Pull-quote SVG in your brand palette.
▦
LinkedIn carousel
Multi-slide deck, 1080×1080.
↗
Chart
Manual data or Polygon stock pull.
$
Stock list
Watchlist → sparkline carousel or thread.
◐
AI image
nano-banana / gpt-image-2, on-brand.
Ready to draft like yourself?
Currently in closed beta. Sign in to continue, or reach out for an invite.