Skip to content

v2.4.0 — Task-appropriate models

Choose a tag to compare

View all tags
@laanito laanito released this 31 Oct 21:11
· 54 commits to main since this release
v2.4.0
e8aaddb
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Task-appropriate models

OpenTerminalUI v2.4.0 gives quick classification and deeper research separate, explicit model profiles.

  • Fast handles headline sentiment, stock emotion and query intent. Existing LLM_* settings remain its configuration.
  • Deep handles Brain answers, comparison briefings, stock briefings, Interrogate, market/collection outlooks, portfolio risk, backtest explanation and filing extraction. Set LLM_DEEP_MODEL to use another model on the existing server, or configure an independent endpoint/key/timeout.
  • All twelve tasks share resolved model identity and bounded request lifecycles. Browser workflows honor announced budgets without automatically replaying interrupted inference. Research and sentiment/emotion caches separate model configurations.
  • Comparison briefings distinguish factual observations from conditional interpretations, possible mechanisms, competing explanations and evidence to investigate. Citations anchor observed inputs; they do not prove mechanisms or certify model reasoning.
  • Experimental filing reports retain model/lexical provenance in an additive companion table, while older records remain honestly provenance-unknown.

Upgrade notes

Existing single-model installations remain supported. Without Deep configuration, deeper tasks use the default model and say so. Once Deep is configured, failures never silently substitute Fast. An omitted Deep URL may inherit the existing endpoint/key; an explicit Deep URL requires its own credentials. Embedding configuration and indexing remain unchanged.

No manual database migration or dependency change is required. The experimental filing provenance table is created lazily without altering existing report columns or catalyst data.

Verification

The maintainer tested and accepted functionality, answer quality, source accuracy and response times through the individual PRs. Local verification passed 1,461 backend tests (one skipped; 76.33% coverage), all 553 frontend tests and fourteen Chrome smoke journeys. Release-preparation PR #239 passed build-and-test, clean-install and postgres-contract before merge.

This release is published from merge commit e8aaddbf78b6d94175620ab12c3dd757b4d643a3. Deployment is managed separately. Host acceptance is not a controlled cross-model benchmark or a guarantee of investment outcomes.

Assets 2
Loading