LiveContext CE v0.2.11
What is new
Upgrading
- A self-hosted install upgrades cleanly again. Two migrations that only ever applied to the hosted edition were not neutralised for CE, so an existing install could refuse to boot after pulling a newer image. They are skipped now, as the rest of the cloud-only schema already was.
Workflow canvas
-
Every button attached to a node is square now. The buttons hanging off a node were the last round islands left on the canvas: pills that grew under the pointer and carried a drop shadow, sitting next to flat square chrome. The bottom bar, the play and rerun buttons, the trigger pin and its launcher, the fleet triggers, delete and duplicate, and the "+" that adds a node from a handle all take the same shape as the rest of the app. One shared style rather than four copies of it, which is what had let the pin button quietly lose its outline.
-
A run can be launched again from an epoch you are reading. Opening one epoch of a run is a record, so the controls under a node step aside. That left no way to fire the workflow again without going back to the cumulative view first, and for a chat, form or webhook trigger no way at all. Every trigger now offers its run button while an epoch is focused: it returns to all epochs, then does exactly what that trigger normally does, fire immediately or open the panel where you type the message or fill the form. The button stands down on the run's newest epoch, where the normal one is already there, and stays away on a finished run, which cannot be fired into.
-
Step-by-step runs can be stepped from the canvas again. Since the run history moved into the side panel every run surface opens through an epoch, and the play and rerun buttons on non-trigger nodes silently vanished with it: no error, badges still drawn, and a run that simply could not be advanced. Reading the live state through its own epoch is not history, so it stays a console; only an older epoch is read-only.
-
See what a run produced without opening every file. A run with a dozen file-producing nodes meant clicking each preview one by one. The canvas toolbar carries a single control that unfolds them all, or folds them back, and it only appears when the canvas actually shows a file.
Run panel
-
The run bar counts the epochs a run has. It was reading the engine's internal cursor, which moves on to the next epoch as soon as one finishes, so a workflow fired once announced "All epochs (2)" directly above a list showing one. The number and the list now come from the same place.
-
The sub-workflow tab has its run history back. The tab opened on one run, the pinned production one, and stopped there: no way back, no way to reach another run of that workflow. Picking a run there now lands, and does it without dragging the page behind the panel along with it. Coming back up to the list, the run you were reading is marked, so "which one was that?" has an answer.
-
Reviewing an approval shows the item you are deciding on. Clicking a pending item opened the review with the variable picker where the item's own data belongs, so the decision was made blind. A node under review now counts as having run data to inspect, and the reviewer sees the input that item arrived with.
Execution
-
A step that reads a value which does not exist now fails, instead of running on emptiness. A parameter pointing at an output no node ever produces used to resolve to nothing and let the step proceed, so the failure surfaced several nodes later as an unrelated error, or not at all. It stops at the step that asked for it, and says what it was looking for.
-
The output names an agent is given are the ones actually written. The builder handed agents keys no mapper ever produces, so a workflow could be wired to a field that never exists at runtime. The documented outputs are aligned with what each node really writes, the Switch node's selected branches included, which had been described as a list while it has always been a string.
-
Data tables read and write in the workspace you are actually in. The table views sent no active-workspace header, so an account belonging to several workspaces could be shown the wrong one's rows.
Sub-workflows
- A sub-workflow that cannot run says what to do about it. The "no active run" message had lost the sentence naming the action that creates one, leaving a dead end where there was a next step.
Under the hood
-
A lighter image and a tidier tree. Unreachable frontend files, dead backend code, an unused 19 MB landing animation and 29 MB of stray build artefacts are gone; two unused dependencies and a duplicate lockfile with them.
-
Every interface string exists in every language. Locale parity is now guarded repo-wide rather than trusted, so a key added in English can no longer ship untranslated to the other five.
Update an existing install
git pull
docker compose pull
docker compose up -d
The compose file pins this release's image tag, so run git pull first. Your data lives in Docker volumes and is kept across updates.
Fresh install
git clone https://github.com/livecontext-ai/livecontext-ce
cd livecontext-ce
docker compose up -d
Then open http://localhost:3000
Images
Pull from whichever registry you prefer, the digests are identical.
-
ghcr.io/livecontext-ai/livecontext-ce:v0.2.11 -
ghcr.io/livecontext-ai/livecontext-ce-frontend:v0.2.11 -
ghcr.io/livecontext-ai/livecontext-ce-bridge:v0.2.11 -
ghcr.io/livecontext-ai/livecontext-ce-screenshot-renderer:v0.2.11(opt-in renderer profile) -
docker.io/livecontext/livecontext-ce:v0.2.11 -
docker.io/livecontext/livecontext-ce-frontend:v0.2.11 -
docker.io/livecontext/livecontext-ce-bridge:v0.2.11 -
docker.io/livecontext/livecontext-ce-screenshot-renderer:v0.2.11(opt-in renderer profile) -
quay.io/livecontext/livecontext-ce:v0.2.11 -
quay.io/livecontext/livecontext-ce-frontend:v0.2.11 -
quay.io/livecontext/livecontext-ce-bridge:v0.2.11 -
quay.io/livecontext/livecontext-ce-screenshot-renderer:v0.2.11(opt-in renderer profile)
Each image is also tagged v0.2, v0 and latest if you prefer to track a moving line rather than pin an exact release.