Skip to content

perf(models): reduce connection and default selection latency - #391

Merged
Angnuo (angnuoli) merged 2 commits into
microsofthackathons:mainfrom
angnuoli:codex/model-settings-latency-20261117
Nov 17, 2026
Merged

Angnuo (angnuoli) merged 2 commits into
microsofthackathons:mainfrom
angnuoli:codex/model-settings-latency-20261117

Conversation

@angnuoli

@angnuoli Angnuo (angnuoli) commented Nov 17, 2026 •

Copy link
Copy Markdown
Collaborator

Description

Saving or removing model connections and switching the default model waited for the full config.apply/config.patch response path even for narrowly scoped changes. Use revision-guarded config.set on the verified OpenClaw 2026.10.10 runtime, then explicitly confirm runtime application before reporting success.

Related Issue

None.

Changes

  • Accelerate Qwen/MiniMax connection-only saves and removals. Retain the original path for reference cleanup, model/plugin edits, unsupported runtimes, or restart modes, and verify that a removed provider is actually absent.
  • Accelerate cloud and prepared-local default-model selection while preserving credentials, fallbacks, aliases, agent overrides, catalog definitions, and model policies. Build complete writes from the unredacted source and retain conflict, availability, local-file, and applied-revision checks.
  • Keep uncertain writes non-repeating and preserve local-selection recovery after a lost response.
  • Add isolated Gateway comparison tests and document the measurements and their limits.
  • Point the MiniMax API-key button to https://platform.minimax.cn/console/plan.
  • Allow existing Copilot worker lifecycle tests enough startup/exit time under parallel Windows test load; authentication behavior is unchanged.

Measured scope

Isolated configuration-application medians (two retained samples per method/provider after warmup):

Default selection config.patch New path
Qwen 6.033 s 2.703 s
MiniMax 6.857 s 2.941 s
Local model configuration 5.786 s 2.545 s

These are not end-to-end UI guarantees: real provider validation, Copilot authentication, MXC attestation, GGUF preparation/loading, and rendering are excluded. Connection-save/removal comparisons, including substantial run-to-run variance, are documented in docs/reference/provider-api-key-save-latency-2026-11-17.md. Default-selection details are in docs/reference/default-model-selection-latency-2026-11-17.md.

Type of Change

  • Bug fix / performance improvement
  • Documentation update

Testing

  • Targeted connection, removal, model-selection, and recovery regression suites passed.
  • Opt-in real Gateway tests passed for provider saves/removals and 18 default selections, using temporary state and simulated credentials/endpoints.
  • Production desktop/renderer build succeeded.
  • Repository lint and formatting checks passed (existing warnings remain; no errors).
  • Full desktop and renderer suites on the upstream-based branch: 2,749 main-process tests passed (59 skipped), 1,159 renderer tests passed (2 skipped), and runtime/resource Node tests passed.
  • Production end-to-end UI latency remeasurement

Checklist

  • Self-review completed.
  • Regression coverage and opt-in reproduction instructions included.
  • Documentation distinguishes confirmed runtime readiness from an early write receipt and isolated timings from production latency.

@angnuoli
Angnuo (angnuoli) merged commit 191d3b3 into microsofthackathons:main Nov 17, 2026
4 checks passed
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.

1 participant