Work items
Settings → Work Items (/admin/settings/ai/work-items) configures the LLM agent that correlates sessions to work items (issues and PRs), plus the scope switches that decide where that agent runs. The agent is off by default and fails open toward “no assignment” — it never writes a wrong high-precedence link.
LLM correlation agent
Section titled “LLM correlation agent”| Key | Type | Default | Notes |
|---|---|---|---|
WorkItems:Correlator:Agent:Enabled | bool | false | Global switch for LLM correlation. |
WorkItems:Correlator:Agent:Provider | enum anthropic / openai / kurrentai / cohere / openrouter | anthropic (runtime) | Form renders blank; an unset value falls back to Anthropic. Reuses the eval-runner credentials from AI Providers. |
WorkItems:Correlator:Agent:Model | string | selected provider’s default judge model (runtime) | Form renders blank; an unset value is post-configured from the effective provider (e.g. Anthropic → claude-sonnet-4-6). |
The global switch above turns the agent on; two per-scope tables decide where it actually runs.
- Projects — shown only on plans that allow projects. A per-project switch; a project’s setting governs all of its repositories.
- Repositories — a per-repository switch.
Enabling either scope backfills the last 30 days of unassigned sessions in that scope.
Related
Section titled “Related”- Work items — what work items are and how sessions attach to them (user guide).
- AI providers — the vendor credentials the correlation agent reuses.
- Embeddings — the embedding pipeline Work Items shares with memories and eval dedup.