fix: preserve thought_signature across parallel tool calls for Gemini 3+

[toubatbrian]

Fixes a 400 error when using Gemini 3+ models with parallel tool calls. The thought_signature is now correctly propagated to all tool calls in a batch instead of only the first one. Also removes a hardcoded parallelToolCalls: true in the voice pipeline that was overriding user configuration.

Summary by CodeRabbit

Release Notes

• Bug Fixes

  • Fixed preservation of context signatures across parallel tool calls for Gemini 3+ models to improve inference quality with the inference gateway.

• Chores

  • Updated configuration to use model options for controlling parallel tool call behavior.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[changeset-bot]

🦋 Changeset detected

Latest commit: 5c1d079

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 18 packages

Name	Type
@livekit/agents	Patch
@livekit/agents-plugin-anam	Patch
@livekit/agents-plugin-baseten	Patch
@livekit/agents-plugin-bey	Patch
@livekit/agents-plugin-cartesia	Patch
@livekit/agents-plugin-deepgram	Patch
@livekit/agents-plugin-elevenlabs	Patch
@livekit/agents-plugin-google	Patch
@livekit/agents-plugin-hedra	Patch
@livekit/agents-plugin-inworld	Patch
@livekit/agents-plugin-livekit	Patch
@livekit/agents-plugin-neuphonic	Patch
@livekit/agents-plugin-openai	Patch
@livekit/agents-plugin-resemble	Patch
@livekit/agents-plugin-rime	Patch
@livekit/agents-plugin-silero	Patch
@livekit/agents-plugins-test	Patch
@livekit/agents-plugin-xai	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR modifies how tool extra data (thought_signature) is preserved during parallel tool calls in Gemini 3+ models, and updates the Agent's LLM configuration to rely on modelOptions for controlling parallel tool call behavior instead of hardcoded settings.

Changes

Cohort / File(s)	Summary
Semantic Release Configuration .changeset/salty-regions-relate.md	New changeset file declaring a patch release for @livekit/agents with a note about preserving thought_signature across parallel tool calls for Gemini 3+.
LLM Tool Call Streaming agents/src/inference/llm.ts	Modified tool-call streaming logic to preserve toolExtra across parallel tool calls within a batch, allowing subsequent tool calls to inherit thought_signature from the first call. toolExtra is conditionally updated only when extra_content exists on the current tool call; otherwise, the previous value is retained. Reset now occurs at the end of the response batch to prevent leakage into subsequent responses.
Voice Agent Configuration agents/src/voice/agent.ts	Removed explicit parallelToolCalls: true setting from LLM chat API calls. Agent now relies on the LLM's modelOptions (e.g., parallel_tool_calls) to control parallelism, with added comments explaining configuration via new inference.LLM({ modelOptions: { parallel_tool_calls: false } }).

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Suggested reviewers

• lukasIO
• theomonnom

Poem

🐰 A thought preserved through whispered calls,
Where Gemini's tools dance through parallel halls,
No hardcoding chains, just options that flow,
The signature passes—let inference go! 🌙

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The PR description is minimal and does not follow the provided template structure with required sections like 'Changes Made' and 'Pre-Review Checklist'.	Expand the description to follow the template format, including detailed 'Changes Made' section and complete pre-review checklist items.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the main fix: preserving thought_signature across parallel tool calls for Gemini 3+, which directly corresponds to the primary change in the PR.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

---

📜 Recent review details

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between ebfb82f and 5c1d079.

📒 Files selected for processing (3)

• .changeset/salty-regions-relate.md
• agents/src/inference/llm.ts
• agents/src/voice/agent.ts

🧰 Additional context used

📓 Path-based instructions (3)

**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (.cursor/rules/agent-core.mdc)

Add SPDX-FileCopyrightText and SPDX-License-Identifier headers to all newly added files with '// SPDX-FileCopyrightText: 2025 LiveKit, Inc.' and '// SPDX-License-Identifier: Apache-2.0'

Files:

• agents/src/inference/llm.ts
• agents/src/voice/agent.ts

**/*.{ts,tsx}?(test|example|spec)

📄 CodeRabbit inference engine (.cursor/rules/agent-core.mdc)

When testing inference LLM, always use full model names from agents/src/inference/models.ts (e.g., 'openai/gpt-4o-mini' instead of 'gpt-4o-mini')

Files:

• agents/src/inference/llm.ts
• agents/src/voice/agent.ts

**/*.{ts,tsx}?(test|example)

📄 CodeRabbit inference engine (.cursor/rules/agent-core.mdc)

Initialize logger before using any LLM functionality with initializeLogger({ pretty: true }) from '@livekit/agents'

Files:

• agents/src/inference/llm.ts
• agents/src/voice/agent.ts

🧠 Learnings (2)

📚 Learning: 2026-01-16T14:33:39.551Z

Learnt from: CR
Repo: livekit/agents-js PR: 0
File: .cursor/rules/agent-core.mdc:0-0
Timestamp: 2026-01-16T14:33:39.551Z
Learning: Applies to **/{examples,test}/**/*.test.ts : Include both basic streaming and tool calling tests to verify full LLM functionality

Applied to files:

• agents/src/inference/llm.ts

📚 Learning: 2026-01-16T14:33:39.551Z

Learnt from: CR
Repo: livekit/agents-js PR: 0
File: .cursor/rules/agent-core.mdc:0-0
Timestamp: 2026-01-16T14:33:39.551Z
Learning: Applies to **/*.{ts,tsx}?(test|example|spec) : When testing inference LLM, always use full model names from `agents/src/inference/models.ts` (e.g., 'openai/gpt-4o-mini' instead of 'gpt-4o-mini')

Applied to files:

• agents/src/inference/llm.ts
• agents/src/voice/agent.ts

🔇 Additional comments (4)

.changeset/salty-regions-relate.md (1)

1-5: Clear changeset entry

The patch release note matches the behavior change and is concise.

agents/src/inference/llm.ts (2)

431-456: Preserving toolExtra across parallel tool calls is consistent with Gemini behavior

Line 436–456 keeps the thought_signature available for subsequent tool calls in the same batch, which fits the stated goal.

---

468-477: Good safeguard resetting toolExtra at response end

Line 475–476 prevents metadata leakage across responses while still supporting parallel batches.

agents/src/voice/agent.ts (1)

331-333: Configurable parallel tool calls is a good default

Line 331–333 makes parallelism follow modelOptions, which is clearer and more flexible for users.

_{✏️ Tip: You can disable this entire section by setting review_details to false in your review settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[toubatbrian]

Note: although this was hardcoded to true, the actual value was still correctly been overriden by user config inside LLM in previous version.

[Hormold]

it works for me