Releases: avifenesh/eigen
Release list
v0.2.5
What's Changed
- fix(gui): smooth scrolling — wheel easing, threaded scroll, anchor on first measure by @avifenesh in #39
- release: v0.2.5 — regression fixes by @avifenesh in #40
- feat(llm): Mixture-of-Agents virtual provider + council author fallbacks by @avifenesh in #41
Full Changelog: v0.2.4...v0.2.5
Contributors
Assets 2
v0.2.4
- Notes view renders markdown — was dumping the raw
.mdsource in a<pre>; now rendered with the shared Markdown component (edit mode keeps the raw textarea). Verified live. - De-flaked
TestPersistAcrossDaemonRestart— waits for the transcript to be durably on disk before the simulated restart, so it no longer races the agent goroutine's write under CI load.
v0.2.3
GUI fixes.
- Theme config now works in the GUI — deepteal / nord / gruvbox apply via
<html data-theme>, live the moment you change it in Config (no restart). The Svelte GUI previously ignored the theme key (TUI-only). - Activity indicator appears immediately on send, so a slow streaming model no longer looks frozen while it thinks before the first token.
- Markdown: memoized lexing (no re-lex/re-render on scroll), tight list items render inline (no stray blank line after
-/1.), block-HTML classified correctly. - Streaming overlap fixed — the in-flight reply no longer renders over the messages above and snap back on commit.
- Committed frontend bundle —
go build/ releases now embed the real GUI; previously the bundle was gitignored so committed frontend fixes never shipped. - Fallback provider forwards streaming + effort/search/fast and surfaces failover (GLM effort picker + streamed thoughts).
v0.2.2
Patch release — fixes GLM (and any fallback-wrapped reasoning model) showing no thoughts and no effort control.
- Fallback wrapper now forwards streaming + capabilities. The daemon wraps the primary provider in a quota-freeze fallback chain; that wrapper previously implemented only Complete, so a wrapped GLM wasn't a Streamer (→ no streamed text/reasoning, no "thoughts") and didn't expose EffortSetter/Searcher/FastModer (→ the GUI hid the effort dropdown, SetEffort no-op). It now implements Stream and forwards effort/search/fast to the active provider.
- Failover is surfaced, never silent. A primary→fallback switch emits a visible note with the cause (" unavailable (…) — falling back to for this turn"), carried per-turn on the context (thread-safe across concurrent sessions).
- Fixed a data race on the fallback freeze timestamp.
Combined with 0.2.1 (GLM reasoning_effort high|max + the seq-reassembly streaming fix), thoughts + the off|high|max effort selector should now appear for GLM-5.2.
v0.2.1
Patch release.
- GUI streaming: the final message, pre-tool preamble, and model "thoughts" now render. The transcript's seq-reassembly latched the window to seq 1, but the pump's per-session seq does not reset on re-attach — so a re-entered chat parked every event until an overflow valve fired, dropping short turns and trailing messages. It now latches from the live pump's actual base (replay-safe).
- GLM-5.2 reasoning: streams reasoning_content by sending reasoning_effort (high|max, per z.ai coding-plan docs) alongside thinking.type; the GUI effort selector now offers off|high|max.
- daemon compactor now rides the quota-failover chain; misc test isolation.
v0.2.0
First tagged release — working-station build-out: connectors (remote-MCP+OAuth, Google, Obsidian, revuto), cross-project work board + kanban, station dream, full Notes/Reviewers views, GPU sparkline; memory subsystem overhaul (canonical git-root scoping, scope merge, any-scope note move, mem_forget decay, scroll fix); quota-freeze failover incl. compaction, per-rule model chains. Default build is webkit-free; Wails GUI via make gui-desktop.