Skip to content

feat(ui): add moderated group conversations - #12051

Draft
localai-org-maint-bot wants to merge 3 commits into
masterfrom
feat/group-chat-11703
Draft

localai-org-maint-bot wants to merge 3 commits into
masterfrom
feat/group-chat-11703

Conversation

@localai-org-maint-bot

Copy link
Copy Markdown
Collaborator

Description

Closes #11703.

Add Group chat to the Chat page so users can moderate one shared conversation between 2–6 model participants. Users can grant individual turns or run 1–10 ordered rounds, with a Stop control. Each request includes the objective and completed contributions with speaker attribution.

The page uses the existing chat API and permission. It keeps text history only while open. Cancelled or failed responses remain marked incomplete and are excluded from later prompts. Documentation covers setup, turn controls, memory behavior, and limits.

Notes for Reviewers

Draft pending browser verification. Local checks passed from core/http/react-ui:

  • node --test tests/group-chat.test.mjs — 15 tests covering attribution, turn order, bounds, malformed/truncated streams, cancellation, and errors.
  • npm run build — includes the GroupChat bundle.
  • npm run lint:inline-styles — baseline unchanged.
  • npx playwright test e2e/group-chat.spec.js --list — 10 browser tests discovered.

Playwright execution is blocked locally: this Alpine environment cannot launch the cached glibc Chromium, and the network proxy rejects browser/dependency downloads. The browser specs are included for CI. Local JS dependencies came from an existing cache because npm registry access was blocked. No real-model inference or full coverage run was performed.

Signed commits

  • Yes, I signed my commits. (Assisted-by attribution included; human sign-off left to the maintainer as instructed.)
  • Documentation updated (docs/content/) for user-facing changes, or not applicable

Let users give installed models individual turns or ordered rounds in a shared conversation. Attribute completed responses and exclude interrupted output from subsequent prompts.

Add streaming and cancellation tests, browser coverage for CI, and a guide for the session-only page.

Assisted-by: Codex:GPT-6
Let model descriptions wrap without losing the install action offscreen.
Give page table rules precedence over the shared table styles.

Separate background facet probes from navigation assertions and wait for
asynchronous size sorting and masonry layout. Fix the chat fixture clock
so runs near midnight keep the expected conversation groups and order.

Assisted-by: Codex:gpt-6
Playwright joins adjacent table-cell text. A llama-cpp backend followed
by 99 requests then matches p99, although no percentile is shown.
Require word boundaries while keeping the absence assertion.

Assisted-by: Codex:gpt-6

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

User-controlled Multi-Agent Conversation (Human-in-the-loop)

1 participant