AFFiNE vs Confluence

AFFiNE publishes a dated canary build most days alongside a slower beta line, and the two carry the same commits at different points on the maturity curve. The recent stream is maintenance-shaped: dependency security bumps in Mermaid and Electron, a journal lookup fix for trashed docs, retry handling for failed transcription jobs. The last directional move was workspace BYOK profiles with model catalogs and key rotation in early August, which sits on top of the MCP credential work from July.

Two threads run in parallel. The AI thread is about who supplies and controls the model - BYOK profiles, MCP credentials, access modes - rather than adding model features. The other thread is self-host ergonomics, and the Render blueprint is the clearest recent instance: AFFiNE keeps lowering the cost of running your own instance while keeping the AI layer pluggable. Between those, the canary absorbs a steady flow of upstream security patches with little product change attached.

Expect the self-host track to keep collecting deploy targets and the BYOK work to extend to more provider surfaces; on current cadence the next 0.27 beta will consolidate the canary work rather than introduce anything new.