Obsidian, Bear, Reflect — and where Carabase fits.
We use Reflect ourselves. These are great tools. Carabase solves a different problem.
Two further axes worth naming up front. Capture surface: Carabase pulls from your calendars, mail, code, browser, photos, messages, and health — through a connector substrate. Obsidian / Bear / Reflect expect you to bring your own content. Agent loop: every read AND write surface is exposed as an MCP tool, plus a typed CLI. The other three either have no agent loop (Bear), a chatbot-in-sidebar (Reflect), or a plugin-driven ecosystem that approximates one (Obsidian).
Feature comparison
The same question, answered very differently.
carabase CLIThe honest take
No hit pieces. Just different tools for different problems.
Incredible plugin ecosystem. Markdown on disk. Full ownership of your files. But it's a note-taking tool — not a knowledge engine. You build the graph manually. AI is bolted on via community plugins, not native to the architecture.
Perfect for quick capture. Gorgeous typography. Feels right on a Mac. But no knowledge graph, no AI, no agent interoperability. It's a writing tool, not a thinking tool.
Beautiful execution. Smart backlinks. GPT-powered assistant. But cloud-hosted (E2E encrypted, their infrastructure), no self-hosting option, no MCP server, no temporal validity on facts, no auditable provenance per claim. Their model knows your notes. Yours.
Your AI doesn't just search your notes — it builds a typed knowledge graph with temporal validity, an associative memory network for sense-making, exposes every read and write as an MCP tool, and lets you review what it inferred before promoting it. Self-hosted, keyboard-first, terminal-fluent, agent-native. Yours.
Early access
Carabase is for people who care where their data lives and want their existing AI tools to understand the context around their work.
The private alpha is open now.