[tang-vu]

Summary

• replace the Chat API Endpoint placeholder with an activatable POST /webhook/ods-chat-completions workflow
• validate and bound messages, roles, model ID, temperature, output tokens, total input size, and non-streaming mode
• forward a normalized request to the local llama-server with a three-minute inference timeout
• emit an allowlisted OpenAI-compatible completion shape or explicit 400/502 error JSON

Why this matters

The dashboard catalog promises a one-minute REST chat endpoint, but enabling the template previously imported only a manual trigger and setup note. This PR turns that reachable catalog action into a usable local API facade with safer limits and predictable responses.

The invariant is: unvalidated fields never reach llama-server, streaming cannot enter a response-node workflow, upstream payloads are not reflected wholesale, and every accepted request has one terminal JSON response.

Overlap check

Searched open and closed upstream PR titles/bodies for n8n chat API endpoint workflow, reviewed the latest 1,000 open PRs and changed production files, and inspected 01-chat-endpoint.json history. No open or closed PR implements this workflow or changes its production file; upstream history only contains the repository rename. Other runnable n8n PRs operate on distinct catalog entries.

The HTTP Request node shape was checked against n8n's official implementation: https://github.com/n8n-io/n8n/blob/master/packages/nodes-base/nodes/HttpRequest/V3/HttpRequestV3.node.ts

Test plan

• python -m pytest test_n8n_chat_endpoint_template.py test_workflows.py -q ? 48 passed
  • authenticated public enable imports the exact shipped graph and activates its n8n ID
  • webhook path, llama request, timeout, non-streaming body, validation ceilings, error branches, and allowlisted output
  • all existing workflow-management tests remain green
• docker run --rm ... n8nio/n8n:2.6.4 import:workflow --input=/workflows/01-chat-endpoint.json ? successfully imported 1 workflow using the exact image ODS ships
• python -m json.tool ods/config/n8n/01-chat-endpoint.json
• python -m compileall -q .../test_n8n_chat_endpoint_template.py
• git diff --check

Design and operational notes

• This workflow intentionally supports text messages and non-streaming completion only; direct ODS APIs remain the path for streaming, tools, and multimodal payloads.
• The existing n8n/llama-server network boundary and dependency gate are reused; no external credential or service is introduced.
• HTTP failures continue into a controlled 502 response. Successful output retains only completion identity, one assistant choice, finish reason, and sanitized token counters.
• Exact-version CLI import proves n8n 2.6.4 accepts the graph. A live inference call was not claimed because no running model endpoint was part of this validation.
• Rollback restores the catalog placeholder with no persisted-data migration.

Generated with Codex.

Batch compatibility

• Recommended merge/application order: feat(installer): add configuration-first TUI #3706 -> feat(test): support multi-service validation subsets #3707 -> feat(api): add conditional model-state reads #3708 -> feat(api): batch model route evidence queries #3709 -> feat(api): add GPU history query controls #3710 -> feat(dashboard): export GPU history as CSV #3711 -> feat(dashboard): focus GPU topology links #3712 -> feat(mobile): bootstrap remote ODS Talk portal #3713 -> feat(workflows): make voice memo runnable #3714 -> feat(workflows): make chat endpoint runnable #3715. The PRs have no runtime dependency; this order also covers the shared MODEL-SWITCHBOARD.md edits in feat(api): add conditional model-state reads #3708/feat(api): batch model route evidence queries #3709.
• Synthetic integration head: tang-vu/DreamServer commit 5e4e354 on integration/features-ten-20260904, built from upstream/main 6ff9b4f plus all ten PR heads.
• Combined validation: 12 BATS tests passed; 132 focused API/workflow pytest tests passed; 6 dashboard Vitest tests passed; ESLint and the 1,771-module Vite production build passed; make lint, installer integration (69 passed, 3 skipped), the 6-test ods-test contract, mobile dispatch smoke, and exact n8n 2.6.4 imports for both workflow templates passed. git diff --check is clean.
• Full-suite limitation: make test reaches tests/test-installer-noninteractive-sudo.sh and reports 3 passed / 2 failed when WSL runs as root. The same two assertions fail unchanged on the clean upstream base 6ff9b4f, so this is recorded as environment/baseline evidence rather than attributed to this batch. All required GitHub checks for the ten PR heads are green as of 2026-09-04.

[Lightheartdevs]

Thanks for this contribution. We're closing outside-contributor pull requests that have had no activity for 14 days or more, so the queue reflects work that can still land. This isn't a judgment on the change itself. If it's still needed, please rebase onto main and reopen it or open a focused PR. See #7253 for details and the contribution policy.