Repository navigation
Releases
Releases · NovemberFalls/plexar-studio
v2.1.57
Full Changelog: v2.1.56...v2.1.57
v2.1.56
v2.1.55
v2.1.54
v2.1.53
Plexar Studio 2.1.53 — agents you can see. Covers everything since 2.1.47.
New
- Agent workers. An agent running in a Studio pane can start its own worker panes: Claude Code, Codex, Plexar Harness or a plain shell, optionally each on its own git worktree. It can prompt them, wait on them, read their answers, send keys, and run commands in a shell worker and wait for a line of output. Workers appear in the sidebar nested under the session that started them, and you can open, watch or type into any of them.
- Limits enforced by Studio: each pane has its own key and can only control its own workers; workers cannot start workers; at most 8 per session; a worker never gets more permission than its parent; an agent cannot answer a question a worker is asking you.
- See Claude's own agents. Agents a Claude Code session runs with its Agent tool are listed under that session with their task, model and a running/done dot. Click one to read its latest output. They clear themselves 30 minutes after finishing.
- Hide a pane without ending it. The new
–button takes a pane off the grid and leaves the session running; the×is now labelled "End session". - A CLI and a guide for agents:
python "$PLEXAR_STUDIO_CLI" --helpinside any pane, andskills/plexar-studio-workers/SKILL.md.
Faster
- Tasks opens quickly. Studio checks Plexar Framework at startup and keeps the Tasks view loaded when you switch away, so coming back no longer reloads it.
Fixed
- Usage and cost land on the right pane. New Claude panes start with their own session id, so a fresh pane can no longer pick up a sibling's transcript or one from a Claude session running outside Studio.
- Panes started when Studio itself was launched from a Claude Code terminal no longer have transcript saving turned off.
- Claude Code's "trust this folder?" prompt shows as waiting instead of idle.
- A Claude pane no longer reads as idle mid-turn, or as WORKING for several seconds after being opened, resized or clicked.
- Shell panes show SHELL instead of a working/idle state.
Docs
/docsand the README cover the agent-worker routes (/api/agent/*, and how they authenticate) and Claude's agents (/api/subagents).
Full detail per version is in CHANGELOG.md.
v2.1.47
Added
- HTTP API docs. The built-in FastAPI docs at
http://127.0.0.1:8420/docsnow group every route by area and carry the app's real version and a security note. See "HTTP API" in the README. - The Tasks view says what it is: a window onto Plexar Framework, our task framework, which isn't public yet and will be open-sourced when it's ready.
Changed
- Plexar models are reached at
https://llm.plexar.tech. The model list shows only chat models: theplexar-signalclassifier is no longer offered. - Plexar Harness panes use your own saved sign-in (
plexar-harness login). Studio no longer requires or passes a shared key unless you save one in Settings. - The default Chat address is
https://chat.plexar.tech.
Fixed
- Plexar Harness panes started with no effort chosen no longer fail every prompt with "unknown reasoning effort".
- The harness model pill and pane headers show the model's name instead of
["plexar","…"], and a saved model the rig no longer serves is marked "not offered" instead of silently used.