v0.2.5 — pack-level connections (You Also Saved)

Your AI now sees the bigger picture across a whole answer. The context pack adds a "You Also Saved" section: related notes already in your brain that your question didn't directly surface — aggregated across every source in the answer, ranked by how many of them connect to it, with anything already shown filtered out. Ask about a topic and your AI can point you to the connected reading you forgot you saved, not just the literal matches. Builds on v0.2.3's per-source connections, lifting them to the level of the whole pack.

ClipBrain v0.2.4 — retrieval-quality polish

Two retrieval-quality polish fixes from fresh-user dogfooding. Context-pack snippets now always show meaningful text instead of occasionally surfacing a bare section heading (like ## Highlights). And a page's connections no longer list the page itself — "Related" links drop self-references, so you only see genuine connections to other captures.

Upgrade: git pull, then restart the ClipBrain server.

ClipBrain v0.2.3 — connections in the context pack

Your AI now sees the connections between your captures. When you ask about something, the context pack includes each source's related notes — links to other things you've saved on the topic — so your AI can point you to related reading you might have forgotten. First step toward a brain that surfaces what you know, not just what you asked for.

Upgrade: git pull, then restart the ClipBrain server.

ClipBrain v0.2.2 — setup always creates a working brain

Fixed a silent setup failure: ./setup.sh could finish with "✅ ready" without actually creating your brain (if no embedding provider was set).

Setup now always creates a working brain — OpenAI embeddings if you have a key, local Ollama if it's running, otherwise a keyword-only brain — verifies it exists, and tells you whether semantic search is on. Captures and search work out of the box now, even without an API key.

Upgrade: git pull && ./setup.sh (safe to re-run).

ClipBrain v0.2.1 — raw highlights

"What did I highlight in X?" now returns your actual highlights, not an AI summary.

The context pack surfaces a page's saved highlights and notes ahead of the derived summary, so asking about a book or article gives you back what you actually marked. Completes the named-source experience from v0.2.0.

Upgrade: git pull, then restart the ClipBrain server (bun run serve or relaunch the launchd service).

ClipBrain v0.2.0 — title-aware retrieval

Ask your AI about a specific book, article, or video by name and get that exact source back.

Queries like "what did I highlight in Deep Work?" or "what did that AI moats video say?" now surface the source you named first, instead of burying it under thematically-similar reading. The context pack adds a title-aware retrieval pass: any captured page whose title or author appears in your query is promoted ahead of the semantic results. Topic and open-ended queries are unchanged.

Upgrade: git pull, then restart the ClipBrain server so it picks up the new server.ts (launchctl kickstart -k gui/$(id -u)/com.clipbrain.serve, or just re-run bun run serve). No re-capture needed — the improvement is at query time.

ClipBrain v0.1.0

Initial public release.

Clip web pages, tweets, Kindle highlights, Gmail threads, YouTube transcripts, and PDFs into a local-first, agent-ready memory layer your AI tools can search over MCP. One shortcut (Cmd+Shift+S) and your AI never starts from zero.

Get started

git clone https://github.com/agentpilled/clipbrain
cd clipbrain && ./setup.sh && bun run doctor

• First-run diagnostics: bun run doctor
• Release validation: bun run launch:check
• Read-only corpus quality + approval-gated cleanup tooling
• Privacy & security docs included; everything runs locally