Releases: ahomsi0/Marven
Releases · ahomsi0/Marven
Release list
v2.9.2
v2.9.1
Highlights
- Agent user message editing — Hover a user bubble in the agent panel, edit the text, and Save & resend to truncate from that turn and rerun the agent (attachments on that turn are kept when possible).
- Groq + images — Multipart vision payloads are sent only on Llama 4 Scout/Maverick; other Groq chat models keep a string
contentand a short note so the API no longer returnsmessages[n].content must be a string. - Agent + images + Lite — When using a Groq vision model with attachments, the stream route forces the standard agent tier, adds an IMAGE INPUT system cue, and preserves image attachments across history turns so models stop claiming they cannot see images.
Run npm run electron:build locally (or your CI) to attach platform installers to this release if needed.
Full Changelog: v2.9.0...v2.9.1
v2.9.0
Highlights
- Agent session persistence — Thread messages, open editor tabs, and file buffers restore when you return to a conversation.
- Disk conflict handling — Workspace files watched on disk; modal when on-disk changes clash with unsaved editor edits.
- Problems tab — LSP diagnostics from open files in the agent workspace.
- Editor polish — Scroll position per tab; command palette and rename wiring improvements.
- Safer links — More reliable
file://and open-external behavior for previews (including Windows paths). - UI fix — Removed the assistant "Edit prompt" control that suggested editing AI replies.
Run npm run electron:build locally to produce installers for your platform, or use your existing CI/build pipeline to attach artifacts to this release if needed.
Full Changelog: v2.8.1...v2.9.0
v2.8.1
v2.8.0
v2.7.0
v2.6.1
Fixes
- AI Backends toggles now reflect in model selector immediately
- Model selector is a fixed size across all tabs and states
Full Changelog: v2.6.0...v2.6.1