Lectern 0.6.0
Lectern 0.6.0 sharpens the brain, makes providers a click to set up, and documents how the whole thing works.
Smarter, leaner memory
Recall used to return the top-scoring files for any message, so a greeting in a broad workspace could surface unrelated files — noise that also risked sending the agent off to read them.
- Relevance floor — a file has to clear a similarity threshold to be recalled. Small talk now recalls nothing; genuine matches still surface.
- Snippets, not whole files — when memory content is actually needed, only the most relevant window of a file enters context. On a typical task that cuts recalled context by roughly 9×.
- The agentic path already passes recalled paths, not file contents, so the agent reads only what it needs. Memory no longer drains tokens.
One-click provider setup
Each provider in Settings (Claude Code, Antigravity, OpenCode, OpenRouter, Ollama) now expands an OS-aware setup panel: the exact install command for your platform, a Copy button, an auth next-step, and links to the full guide and the provider's official page. For the safe user-space installers (OpenCode, Ollama) there's a one-click Install that runs it and streams the output. The one-click path only runs vetted, hardcoded commands.
Documentation
New feature-level docs at getlectern.vercel.app/docs: chat commands and skill replay, the Conductor and how model routing works, the brain and recall, scheduling, and creating/importing/sharing Hub skills.
Fixes
schedule cancelaccepts the short id thatschedule listprints, and reports a real miss instead of silently succeeding.- Recall no longer indexes graphify's own cache into results.
Install
- Linux —
Lectern_0.6.0_amd64.AppImage(chmod +xthen run) orLectern_0.6.0_amd64.deb. - macOS —
Lectern_0.6.0_aarch64.dmg(unsigned; right-click → Open the first time). - Windows —
Lectern_0.6.0_x64-setup.exe(unsigned; SmartScreen → More info → Run anyway). - CLI + daemon —
lectern-cli-linux-x64.tar.gz(lectern,lecternd).
Verify downloads against SHA256SUMS.txt. Everything builds from source under Apache-2.0.