[EtanHey]

Problem

Fixes #31767. Selecting a relay-connected environment for microphone audio fails immediately with “No relay authorization service is available for the environment request.” Speech requests run on the generic HTTP runtime, which lacks the connection session's remote authorizer.

Change

Run web/desktop speech effects through the existing connection atom runtime and application registry. This shares the session authorizer and token renewal across status probes, model management, streaming tickets, uploads, and transcript cleanup. Existing abort signals are forwarded to the effect runner.

Scope and approval

Small, focused fix for an obvious bug in the existing remote dictation capability; no new feature, setting, permission, contract, or visual behavior. The issue above records the desktop trace evidence. Mobile already uses connection-runtime speech atoms and is unchanged.

Verification

• Failing first: the new browser regression, using the real speech HTTP effect and runtime, failed with the exact missing-service error. After the fix, a signed-out relay request reaches the connection authorizer and reports its expected authorization failure.
• vp test run apps/web/src/speech/runtime.browser.test.ts apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx: 21 passed.
• vp run check:changed: format, lint, affected-package typecheck, both knip checks, focused tests, and React compiler checks passed; typecheck emitted existing schema-brand suggestions.
• Local CodeRabbit review exceeded its 45-second budget and was stopped.
• Signed-in relay audio capture/upload and the installed desktop app have not been verified. No installed-app changes or live database writes.

Implementer: Codex. Inner review: Codex approved at d859363, 1 completed round, no blocking findings (user explicitly selected Codex after Claude hit its API rate limit). All latest-head CI checks passed or skipped. Signed-in live audio remains unverified. Regression guard: relay speech request browser test.

🤖 gpt-6.1-sol via T3 Code / Codex harness

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

Speech requests now use a local runtime wrapper that runs HTTP effects through the connection atom runtime. The browser voice input and environment speech hook use this wrapper. A browser test checks the authorization error for an expired DPoP credential when no account is signed in.

Changes: Speech runtime authorization

Layer / File(s)	Summary
Connection-aware runtime and authorization test apps/web/src/speech/runtime.ts, apps/web/src/speech/runtime.browser.test.ts	Adds speechRuntime.runPromise, which runs HTTP effects through the connection atom runtime and forwards optional abort-signal options. A browser test checks that an environment speech status request rejects with the expected authorization error for expired DPoP credentials when no account is signed in.
Speech caller integration apps/web/src/speech/browserVoiceInput.ts, apps/web/src/speech/useEnvironmentSpeechInput.ts, apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx	Updates both speech callers to import the local speechRuntime wrapper. Updates the hook test mock to use the local runtime module.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Medium

Possibly related PRs

• pingdotgg/t3code#30623: Adds the environment dictation hook and browser transcription path that this change updates to use the connection runtime.
• pingdotgg/t3code#31769: Adds cloud transcription and uses authenticated environment requests for speech effects.
• pingdotgg/t3code#30624: Uses useEnvironmentSpeechInput for desktop annotation dictation, a hook updated by this change.

Suggested reviewers: juliusmarminge

Merge Risk: Minimal · up to d8593

The change is mergeable after normal checks. Signed-in relay capture and installed desktop behavior remain unverified, but no specific failure was established.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly describes the relay authorization fix for microphone destinations.
Description check		The description covers the problem, change, scope, and verification. It identifies signed-in audio and the installed desktop app as unverified.
Linked Issues check		[#31767] Speech requests now use the connection atom runtime and application registry. This provides the session authorizer instead of the generic HTTP runtime. The browser regression confirms that an…
Out of Scope Changes check		[#31767] The runtime wrapper, speech call-site changes, and browser tests support the relay authorization fix. The diff shows no unrelated changes.

Full details: Linked Issues check

Explanation

[#31767] Speech requests now use the connection atom runtime and application registry. This provides the session authorizer instead of the generic HTTP runtime. The browser regression confirms that an expired relay request reaches the authorizer rather than failing for a missing service. Signed-in live audio remains unverified.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: c52df24
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 3fe53248-46ba-41eb-a1eb-6f60b4498f77

Commits

Reviewing files that changed from the base of the PR and between c52df24 and d859363.

Files selected for processing: 5

• apps/web/src/speech/browserVoiceInput.ts
• apps/web/src/speech/runtime.browser.test.ts
• apps/web/src/speech/runtime.ts
• apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx
• apps/web/src/speech/useEnvironmentSpeechInput.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.

[juliusmarminge]

Thanks @EtanHey for tracking this down: running speech requests through the connection runtime is the right fix, and your regression test shows it clearly. We're closing this in favour of https://git.cafe/pingdotgg/t3code/pulls/31778, which fixes the same issue (https://git.cafe/pingdotgg/t3code/issues/31767) the same way and also moves the Voice settings panel's own requests (status, model list, download/select/remove) off the generic runtime. On main those still call lib/runtime directly, so choosing a relay destination in Settings would keep failing without that part. Your PR helped confirm the cause.