# Brainfeed Agents read the newsletters. You keep the best ideas. Brainfeed compares every idea to your second brain and sends you only what is new. Swipe to accept. Your agents get smarter with every swipe. Free early access. Your notes stay in your own repo. Works with Claude Code, Codex, Obsidian and GitHub. - Join early access: https://brainfeed.atharvashah.com/#join - Scan your notes (works today): https://brainfeed.atharvashah.com/#try ## Teach Your Second Brain Something New Every Week Your second brain learns from expert newsletters. Swipe to keep the ideas worth knowing. Each card shows the source, the idea, why you see it, and how it relates to your notes: builds on a note, new to your second brain, or conflicts with a note. Facts: 41 newsletters ranked against your notes. 40 notes read in your browser. 5 minutes from repo to first swipe. 0 servers that see your notes. ## Go From Repo to First Swipe in Five Minutes 1. **Connect your notes.** Paste a repo. Brainfeed reads your topics and nothing else. 2. **Pick your sources.** We rank 41 newsletters against your notes. You keep the ones you want. 3. **Swipe every Monday.** Accept, edit or reject. Only accepted ideas reach your notes. ## Nothing Becomes Trusted Until You Say So Agents do the reading and the comparing. You make the one call that matters. Newsletters (sources you trust) → Agents compare (new, builds on, conflicts) → You swipe (accept, edit, reject) → Second brain (trusted notes) → Your agents (work with better context). Every swipe also updates what agents surface next week. ## Same Issue, Very Different Result One Every Frontier issue. Two ways to handle it. **Summarizer:** This week the Every Frontier team shared eight experiments on agents, writing and video. Highlights include a fake Slack for testing agents, a research agent that maps 104 paths, and a storyboard-first video editor. A shorter issue. You still decide what matters, and your notes stay the same. **Brainfeed:** - Test agents against a fake Slack. Write the fake service as plain code so agents run without the real one. (Builds on agent-testing) - Map the research before you run it. A parent agent maps 104 paths, picks five and runs each as its own session. (New to your second brain) - Fast models edit sentences well. Short paragraph edits work. Whole-draft editing still favors larger models. (Conflicts with one-model-for-edits) Three ideas, each checked against your notes. Your swipe decides what your agents learn. ## Every Swipe Teaches Your Agents - **Accept:** The idea becomes a note. Similar ideas rank higher next week. - **Reject:** The source and topic weigh less. You can say why. - **Plain Markdown:** Preferences live in your repo, so you own them and your agents read them. ## Pick Your Way In Start with a prompt or the CLI today. Both tie back to your dashboard. The MCP server comes with early access. - Setup prompt (live): paste it into Claude Code, Codex or any agent. It tells the agent to read each source every Monday and write one note per issue with three claims, one thing to try and links to related notes. - CLI (live): `npx github:HighnessAtharva/brainfeed login`, then `init `, `queue` and `pull`. It syncs with your Brainfeed account. - MCP (preview, not released): ask what you accepted this week, plan experiments from it, save the plan to your second brain. ## Brainfeed Closes the Loop Other Tools Leave Open Summarizers give you less to read. Triage helps you ignore more. Brainfeed gives your agents better context. | Tool | Compares to your second brain | You approve before it is trusted | Feeds your agents | |---|---|---|---| | Brainfeed | Yes | Yes, by swipe | Yes | | Readwise Reader | Limited | No | Through export | | Newsletter digests | No | No | No | | Inbox triage | No | No | No | | Obsidian or Notion with AI | When you ask | Manual | Partly | | Autonomous agents | No | No | Yes, with noise | Based on public product descriptions in November 2026. ## Start With the Newsletters Your Notes Are Missing Paste a GitHub repo. Your browser reads up to 40 notes and ranks 41 newsletters against them. This part works today. Private repo? Use a fine-grained token with read-only access to Contents for this one repo. It stays in the browser tab and goes only to api.github.com. ## Questions You Will Ask ### Is this another summarizer? No. A summarizer stops at shorter text. Brainfeed compares each idea to your second brain, waits for your accept, and gives the result to your agents. ### I already use Readwise and Obsidian. Do I need this? They are good inputs. Brainfeed sits on top and turns their output, or the raw sources, into judged notes your agents can use. ### Won't the agents add noise? Low-quality ideas never enter your notes, because each one needs your accept. Over time the agents learn what you accept and show you less of the rest. ### Will this create more work? The queue is short and ranked. You see only ideas that score as new or relevant against your notes, and you can defer the rest. ### What can I use today? The repo scan and the setup prompt work now. The swipe queue, the source reader and preference learning are in early access. ### Is my private repo safe? The scan runs in your browser. Your token goes to api.github.com and nowhere else, and Brainfeed has no server that sees your notes. ## Join Early Access Teach your second brain something new every week. Join early access. Start with your own newsletters and your own notes. Built by [Atharva Shah](https://atharvashah.com). Source: https://github.com/HighnessAtharva/brainfeed