fix: apply 193 confirmed findings from full-codebase review

[FrancescoRosciano]

Summary

• Rebuilds this PR on top of current main so it ships only the genuine fixes from the full-codebase review, with zero regressions of features that landed on main after the original branch was cut.
• Closes the one-sided Python↔TypeScript parity gaps the review left open and corrects the defects it introduced.
• Drops the internal review artifacts and the unauthorized model-default change that were in the original branch.

Implementation

• Strategy: branched from origin/main; ported the review's real fixes (the 92 files only this PR touched applied cleanly; the 20 files main also changed were manually merged with main as the base), then rebased onto the latest main.
• Preserved from main (a naïve merge would have silently deleted these): call(wait=True)/CallResult (feat: completion-aware outbound calls — call(wait=True) → CallResult + guaranteed teardown #120), built-in transfer_call/end_call injection in pipeline mode (fix(pipeline): inject built-in transfer_call / end_call tools into LLMLoop #115/fix(typescript): pipeline built-in tools + docs events parity for #114/#115 + CHANGELOG backfill #118), Plivo carrier (feat(plivo): add Plivo as a third carrier with full Twilio/Telnyx parity #121/chore(plivo): close remaining gaps — dashboard UI, lib READMEs, packaging keywords #123), is_ws_alive websocket-liveness (fix(websockets): use is_ws_alive across modern websockets>=12 API #113), the pipeline metrics re-arm (fix(metrics): re-arm turn guard in anchorUserSpeechStart — pipeline turns emit no metrics/transcript #125), and the security scrub (chore(security): scrub production Call SID and competitor lineage from public files #129).
• Genuine fixes brought in: DeepFilterNet suppression no-op → 48 kHz resampler; parallel tool-call 400 on Anthropic + Gemini (both SDKs); MCP SSRF guard (both SDKs); off-loop blocking calls (Krisp/Whisper); WS read timeouts + background-task exception surfacing; AMD per-callSid map; ElevenLabs API key made private; resource leaks.
• Parity gaps closed: Gemini merge (Py), MCP SSRF (Py), AMD map (Py), FallbackLLMProvider.cancel_event (Py), Telnyx STT cost (TS), PatterConfigError (TS), Azure cached-token savings (TS), Anthropic cache-token billing (TS).
• Defects fixed: done_callback cancelled-guards, DTMF_EVENTS ordering, OpenAI TTS default restored, EOU emit guard aligned, orphaned conversation-history turn, Telnyx STT logger namespace.
• Default Agent.model aligned across SDKs to gpt-realtime-mini (matches the TS adapter default; same model family / pricing).
• Internal review docs (docs/reviews/*, docs/DEVLOG.md) and the original unauthorized model-default change are not included.

Breaking change?

No runtime breaking change. Two notes for callers:

• Public config collections are now immutable (frozen tuples / readonly arrays) — code comparing them against a plain list/array must compare against a tuple/readonly array.
• Agent.model default string changed gpt-4o-mini-realtime-preview → gpt-realtime-mini (aliases for the same model, identical pricing).

Test plan

• Python: pytest tests/ — 1940 passed
• TypeScript: npm test (1555 passed) + npm run lint (clean) + npm run build (success)
• dashboard-app: tsc --noEmit clean
• /parity-check — no residual public-surface gaps (one internal-flag TTFB divergence noted as optional follow-up)
• No PII / competitor / secret reintroduced (the chore(security): scrub production Call SID and competitor lineage from public files #129 scrub is preserved)

Docs updates

• N/A — CHANGELOG.md ## Unreleased updated; no docs/**.mdx pages touched.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
patter-06b046ce	🟢 Ready	View Preview	May 30, 2026, 4:22 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.